{"meta":{"generated_at":"2026-10-08T14:09:53Z","retrieved_at":"2026-10-08","harnesses":["artificial-analysis","published","vals-ai"],"counts":{"models":401,"benchmarks":147,"measured":4892,"estimated":16443,"mappings":1179,"cross_mappings":6784,"cross_multi_mappings":117,"confidence":{"high":2523,"medium":5185,"low":8735}},"quality_gate":{"min_pairs":5,"min_r2":0.3,"max_loo_pp":15,"n_candidates":6},"confidence_levels":{"high_max_pp":5,"medium_max_pp":10,"min_reliable_n":8,"min_informative_r2":0.5},"dense":{"min_models":8,"min_benchmarks":3},"methods":{"linear":"linear","mm":"Michaelis–Menten","mm_offset":"Michaelis–Menten + offset","mm_offset_inv":"inverse Michaelis–Menten","hill":"Hill","logistic":"offset logistic","linear_mv":"ridge regression (multivariate)","mm_mv":"multivariate Michaelis–Menten","enet_mv":"elastic net (multivariate)"},"providers":{"anthropic":"Anthropic","openai":"OpenAI","google":"Google","xai":"xAI","deepseek":"DeepSeek","zhipu":"Zhipu (GLM)","alibaba":"Alibaba (Qwen)","moonshot":"Moonshot","minimax":"MiniMax","mistral":"Mistral","nvidia":"NVIDIA","meta":"Meta","xiaomi":"Xiaomi","stepfun":"StepFun","other":"Other"},"home":"terminal-bench-4/current"},"capabilities":[{"id":"agentic-terminal","label":"Agentic · terminal"},{"id":"agentic-tool","label":"Agentic · tools"},{"id":"coding","label":"Coding"},{"id":"math","label":"Math"},{"id":"knowledge","label":"Knowledge & reasoning"},{"id":"instruction-following","label":"Instruction following"},{"id":"multilingual","label":"Multilingual"},{"id":"vision","label":"Vision & documents"},{"id":"long-context","label":"Long context"},{"id":"general","label":"General"}],"benchmarks":[{"id":90,"key":"aa-terminal-bench21/current","label":"AA Terminal-Bench 2.1","capability":"agentic-terminal","harness":"artificial-analysis","source_url":"https://benchlm.ai/benchmarks/terminalbench21","featured":true,"n_measured":12,"n_estimated":92,"listed":true,"dense":true},{"id":91,"key":"aa-terminal-bench4/current","label":"AA Terminal-Bench 4.0","capability":"agentic-terminal","harness":"artificial-analysis","source_url":"https://benchlm.ai/benchmarks/terminal-bench-4","featured":true,"n_measured":24,"n_estimated":84,"listed":true,"dense":true},{"id":141,"key":"terminal-bench-2/current","label":"Terminal-Bench 2.0","capability":"agentic-terminal","harness":"published","source_url":"https://benchlm.ai/benchmarks/terminal-bench-2","featured":false,"n_measured":41,"n_estimated":43,"listed":true,"dense":true},{"id":74,"key":"terminal-bench-3/current","label":"Terminal-Bench 3.0","capability":"agentic-terminal","harness":"published","source_url":"https://benchlm.ai/benchmarks/terminal-bench-3","featured":false,"n_measured":13,"n_estimated":83,"listed":true,"dense":true},{"id":103,"key":"terminal-bench-4/current","label":"Terminal-Bench 4.0","capability":"agentic-terminal","harness":"published","source_url":"https://benchlm.ai/benchmarks/terminal-bench-4","featured":false,"n_measured":17,"n_estimated":13,"listed":true,"dense":true},{"id":79,"key":"terminal-bench-hard/current","label":"Terminal-Bench Hard","capability":"agentic-terminal","harness":"artificial-analysis","source_url":"https://benchlm.ai/benchmarks/terminal-bench-hard","featured":true,"n_measured":5,"n_estimated":0,"listed":false,"dense":false},{"id":104,"key":"terminal-bench-science/current","label":"Terminal-Bench-Science 0.1","capability":"agentic-terminal","harness":"published","source_url":"https://benchlm.ai/benchmarks/terminal-bench-science","featured":true,"n_measured":5,"n_estimated":19,"listed":true,"dense":false},{"id":250,"key":"terminal-bench-science-6x-verifier/current","label":"Terminal-Bench-Science 0.1 (6x verifier timeout)","capability":"agentic-terminal","harness":"published","source_url":"https://benchlm.ai/benchmarks/terminal-bench-science-6x-verifier","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":220,"key":"terminal-bench3/current","label":"Terminal-Bench 3","capability":"agentic-terminal","harness":"published","source_url":"https://benchlm.ai/benchmarks","featured":false,"n_measured":2,"n_estimated":0,"listed":false,"dense":false},{"id":20,"key":"terminalbench21/current","label":"Terminal-Bench 2.1","capability":"agentic-terminal","harness":"published","source_url":"https://benchlm.ai/benchmarks/terminalbench21","featured":false,"n_measured":64,"n_estimated":40,"listed":true,"dense":true},{"id":84,"key":"vals-terminal-bench21/current","label":"Terminal-Bench 2.1 (Vals AI)","capability":"agentic-terminal","harness":"vals-ai","source_url":"https://benchlm.ai/benchmarks/terminalbench21","featured":false,"n_measured":61,"n_estimated":70,"listed":true,"dense":true},{"id":55,"key":"aa-agentic-index/current","label":"AA Agentic Index","capability":"agentic-tool","harness":"artificial-analysis","source_url":"https://benchlm.ai/benchmarks/aaagenticindex","featured":false,"n_measured":72,"n_estimated":151,"listed":true,"dense":true},{"id":64,"key":"aa-analyst-agent/current","label":"AA-AnalystAgent","capability":"agentic-tool","harness":"artificial-analysis","source_url":"https://benchlm.ai/benchmarks/aaanalystagent","featured":true,"n_measured":15,"n_estimated":131,"listed":true,"dense":true},{"id":86,"key":"aa-automation-bench/current","label":"AA AutomationBench","capability":"agentic-tool","harness":"artificial-analysis","source_url":"https://benchlm.ai/benchmarks/aaautomationbench","featured":true,"n_measured":25,"n_estimated":135,"listed":true,"dense":true},{"id":65,"key":"aa-enterprise-ops-gym/current","label":"AA EnterpriseOps-Gym","capability":"agentic-tool","harness":"artificial-analysis","source_url":"https://benchlm.ai/benchmarks/aaenterpriseopsgym","featured":true,"n_measured":13,"n_estimated":196,"listed":true,"dense":true},{"id":87,"key":"aa-gdp-pdf/current","label":"GDP.pdf","capability":"agentic-tool","harness":"artificial-analysis","source_url":"https://benchlm.ai/benchmarks/aagdppdf","featured":true,"n_measured":24,"n_estimated":133,"listed":true,"dense":true},{"id":66,"key":"aa-harvey-lab/current","label":"AA Harvey LAB","capability":"agentic-tool","harness":"artificial-analysis","source_url":"https://benchlm.ai/benchmarks/aaharveylab","featured":true,"n_measured":12,"n_estimated":99,"listed":true,"dense":true},{"id":88,"key":"aa-itbench/current","label":"AA ITBench","capability":"agentic-tool","harness":"artificial-analysis","source_url":"https://benchlm.ai/benchmarks/aaitbench","featured":true,"n_measured":17,"n_estimated":159,"listed":true,"dense":true},{"id":89,"key":"aa-tau3-banking/current","label":"AA Tau3 Banking","capability":"agentic-tool","harness":"artificial-analysis","source_url":"https://benchlm.ai/benchmarks/aatau3banking","featured":true,"n_measured":14,"n_estimated":135,"listed":true,"dense":true},{"id":216,"key":"agentslastexam/current","label":"Agents' Last Exam","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/agentslastexam","featured":false,"n_measured":16,"n_estimated":48,"listed":true,"dense":true},{"id":276,"key":"androidworld/current","label":"AndroidWorld","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/androidworld","featured":false,"n_measured":9,"n_estimated":0,"listed":false,"dense":false},{"id":32,"key":"apexagents/current","label":"APEX-Agents","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/apexagents","featured":false,"n_measured":10,"n_estimated":125,"listed":true,"dense":true},{"id":33,"key":"apexagentsaa/current","label":"APEX-Agents-AA","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/apexagentsaa","featured":true,"n_measured":28,"n_estimated":189,"listed":true,"dense":true},{"id":68,"key":"apprenticebench/current","label":"ApprenticeBench","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/apprenticebench","featured":false,"n_measured":18,"n_estimated":94,"listed":true,"dense":true},{"id":38,"key":"automationbench/current","label":"AutomationBench","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/automationbench","featured":false,"n_measured":22,"n_estimated":128,"listed":true,"dense":true},{"id":206,"key":"automationbenchzapier106/current","label":"AutomationBench (Zapier 1.0.6)","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/automationbenchzapier106","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":277,"key":"bankertoolbench/current","label":"BankerToolBench","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/bankertoolbench","featured":false,"n_measured":2,"n_estimated":0,"listed":false,"dense":false},{"id":39,"key":"bfcl-v4/current","label":"BFCL v4","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/bfcl-v4","featured":false,"n_measured":22,"n_estimated":187,"listed":true,"dense":true},{"id":356,"key":"bfclv3/current","label":"BFCL v3","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/bfclv3","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":21,"key":"browsecomp/current","label":"BrowseComp","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/browsecomp","featured":false,"n_measured":45,"n_estimated":143,"listed":true,"dense":true},{"id":123,"key":"claw-eval/current","label":"Claw-Eval","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/claw-eval","featured":false,"n_measured":39,"n_estimated":116,"listed":true,"dense":true},{"id":348,"key":"coworkbench/current","label":"CoWorkBench","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/coworkbench","featured":false,"n_measured":4,"n_estimated":0,"listed":false,"dense":false},{"id":352,"key":"ctirealm/current","label":"CTI-REALM","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/ctirealm","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":94,"key":"cwe-bench-v1/current","label":"CWE-bench v1","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/cwe-bench-v1","featured":false,"n_measured":11,"n_estimated":114,"listed":true,"dense":true},{"id":247,"key":"cwebench/current","label":"CWE-Bench","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/cwebench","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":331,"key":"cybench/current","label":"Cybench","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/cybench","featured":false,"n_measured":2,"n_estimated":0,"listed":false,"dense":false},{"id":40,"key":"cybergym/current","label":"CyberGym","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/cybergym","featured":false,"n_measured":28,"n_estimated":151,"listed":true,"dense":true},{"id":305,"key":"deckbench/current","label":"DECK-Bench","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/deckbench","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":124,"key":"deepplanning/current","label":"DeepPlanning","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/deepplanning","featured":false,"n_measured":7,"n_estimated":50,"listed":true,"dense":false},{"id":35,"key":"deepsearchqa/current","label":"DeepSearchQA","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/deepsearchqa","featured":false,"n_measured":19,"n_estimated":148,"listed":true,"dense":true},{"id":168,"key":"draco/current","label":"DRACO","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/draco","featured":false,"n_measured":6,"n_estimated":117,"listed":true,"dense":false},{"id":118,"key":"exploitgym/current","label":"ExploitGym","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/exploitgym","featured":false,"n_measured":15,"n_estimated":107,"listed":true,"dense":true},{"id":160,"key":"financeagentv2/current","label":"Finance Agent v2","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/financeagentv2","featured":false,"n_measured":8,"n_estimated":0,"listed":false,"dense":false},{"id":27,"key":"gaia/current","label":"GAIA","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/gaia","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":10,"key":"gdpvalaanormalized/current","label":"GDPval-AA","capability":"agentic-tool","harness":"artificial-analysis","source_url":"https://benchlm.ai/benchmarks/gdpvalaanormalized","featured":false,"n_measured":122,"n_estimated":116,"listed":true,"dense":true},{"id":326,"key":"gdpvalrubrics/current","label":"GDPval rubrics","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/gdpvalrubrics","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":63,"key":"gertlabs/current","label":"Gert Labs","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/gertlabs","featured":false,"n_measured":52,"n_estimated":133,"listed":true,"dense":true},{"id":22,"key":"hlewithtools/current","label":"HLE w/ tools","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/hlewithtools","featured":false,"n_measured":22,"n_estimated":131,"listed":true,"dense":true},{"id":41,"key":"jobbench/current","label":"JobBench","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/jobbench","featured":false,"n_measured":30,"n_estimated":150,"listed":true,"dense":true},{"id":300,"key":"kimiclaw247/current","label":"Kimi Claw 24/7","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/kimiclaw247","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":179,"key":"legalagentbenchallpass/current","label":"LAB all-pass (Anthropic harness)","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/legalagentbenchallpass","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":180,"key":"legalagentbenchcriterionpass/current","label":"LAB criterion-pass (Anthropic harness)","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/legalagentbenchcriterionpass","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":181,"key":"legalagentbenchheldoutallpass/current","label":"LAB all-pass (Harvey held-out)","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/legalagentbenchheldoutallpass","featured":false,"n_measured":3,"n_estimated":0,"listed":false,"dense":false},{"id":182,"key":"legalagentbenchheldoutcriterionpass/current","label":"LAB criterion-pass (Harvey held-out)","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/legalagentbenchheldoutcriterionpass","featured":false,"n_measured":3,"n_estimated":0,"listed":false,"dense":false},{"id":129,"key":"mcpatlas/current","label":"MCP Atlas","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/mcpatlas","featured":false,"n_measured":39,"n_estimated":151,"listed":true,"dense":true},{"id":183,"key":"mcpatlasclaimcoverage/current","label":"MCP-Atlas claim coverage","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/mcpatlasclaimcoverage","featured":false,"n_measured":2,"n_estimated":0,"listed":false,"dense":false},{"id":302,"key":"mcpmarkverified/current","label":"MCP Mark Verified","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/mcpmarkverified","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":130,"key":"mcptasks/current","label":"MCP-Tasks","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/mcptasks","featured":false,"n_measured":5,"n_estimated":161,"listed":true,"dense":false},{"id":28,"key":"mle-bench-lite/current","label":"MLE-Bench Lite","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/mle-bench-lite","featured":false,"n_measured":3,"n_estimated":0,"listed":false,"dense":false},{"id":322,"key":"mmclawbench/current","label":"MM-ClawBench","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/mmclawbench","featured":false,"n_measured":2,"n_estimated":0,"listed":false,"dense":false},{"id":350,"key":"mobileworld/current","label":"MobileWorld","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/mobileworld","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":185,"key":"multi-agent-browse-comp-prerelease/current","label":"BrowseComp (10-agent, prerelease)","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/browsecomp","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":137,"key":"osworld/current","label":"OSWorld","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/osworld","featured":false,"n_measured":2,"n_estimated":0,"listed":false,"dense":false},{"id":78,"key":"osworld-verified/current","label":"OSWorld-Verified","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/osworld-verified","featured":false,"n_measured":34,"n_estimated":143,"listed":true,"dense":true},{"id":98,"key":"osworld2/current","label":"OSWorld 2.0","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/osworld2","featured":false,"n_measured":23,"n_estimated":99,"listed":true,"dense":true},{"id":311,"key":"pinchbench/current","label":"PinchBench","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/pinchbench","featured":false,"n_measured":6,"n_estimated":0,"listed":false,"dense":false},{"id":138,"key":"qwenclawbench/current","label":"QwenClawBench","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/qwenclawbench","featured":false,"n_measured":10,"n_estimated":128,"listed":true,"dense":true},{"id":153,"key":"researchclawbench/current","label":"ResearchClawBench","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/researchclawbench","featured":false,"n_measured":19,"n_estimated":154,"listed":true,"dense":true},{"id":42,"key":"skillsbench/current","label":"SkillsBench","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/skillsbench","featured":false,"n_measured":7,"n_estimated":10,"listed":true,"dense":false},{"id":309,"key":"spreadsheetbench2/current","label":"SpreadsheetBench 2","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/spreadsheetbench2","featured":false,"n_measured":2,"n_estimated":0,"listed":false,"dense":false},{"id":29,"key":"tau2-bench/current","label":"τ²-bench results","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/tau2-bench","featured":true,"n_measured":133,"n_estimated":55,"listed":true,"dense":true},{"id":235,"key":"tau2airline/current","label":"τ²-bench Airline","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/tau2airline","featured":false,"n_measured":2,"n_estimated":0,"listed":false,"dense":false},{"id":44,"key":"tau3-bench/current","label":"τ³-bench results","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/tau3-bench","featured":false,"n_measured":18,"n_estimated":74,"listed":true,"dense":true},{"id":142,"key":"toolathlon/current","label":"Toolathlon","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/toolathlon","featured":false,"n_measured":22,"n_estimated":154,"listed":true,"dense":true},{"id":105,"key":"toolathlonverified/current","label":"Toolathlon-Verified","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/toolathlonverified","featured":false,"n_measured":21,"n_estimated":143,"listed":true,"dense":true},{"id":106,"key":"toolathlonverifiedpass3/current","label":"Toolathlon Verified Pass@3","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/toolathlonverifiedpass3","featured":false,"n_measured":4,"n_estimated":0,"listed":false,"dense":false},{"id":107,"key":"toolathlonverifiedpass3all/current","label":"Toolathlon Verified Pass³","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/toolathlonverifiedpass3all","featured":false,"n_measured":4,"n_estimated":0,"listed":false,"dense":false},{"id":25,"key":"vitabench/current","label":"VITA-Bench","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/vitabench","featured":false,"n_measured":12,"n_estimated":71,"listed":true,"dense":true},{"id":332,"key":"webarena-verified/current","label":"WebArena-Verified","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/webarena-verified","featured":false,"n_measured":3,"n_estimated":0,"listed":false,"dense":false},{"id":236,"key":"webvoyager/current","label":"WebVoyager","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/webvoyager","featured":false,"n_measured":2,"n_estimated":0,"listed":false,"dense":false},{"id":45,"key":"wideresearch/current","label":"WideResearch","capability":"agentic-tool","harness":"published","source_url":"https://benchlm.ai/benchmarks/wideresearch","featured":false,"n_measured":15,"n_estimated":155,"listed":true,"dense":true},{"id":30,"key":"aa-coding-index/current","label":"AA Coding Index","capability":"coding","harness":"artificial-analysis","source_url":"https://benchlm.ai/benchmarks/aacodingindex","featured":false,"n_measured":102,"n_estimated":117,"listed":true,"dense":true},{"id":248,"key":"aa-live-code-bench/current","label":"AA LiveCodeBench","capability":"coding","harness":"artificial-analysis","source_url":"https://benchlm.ai/benchmarks/aalivecodebench","featured":true,"n_measured":2,"n_estimated":0,"listed":false,"dense":false},{"id":3,"key":"aa-sci-code/current","label":"AA-SciCode","capability":"coding","harness":"artificial-analysis","source_url":"https://benchlm.ai/benchmarks/aascicode","featured":true,"n_measured":105,"n_estimated":112,"listed":true,"dense":true},{"id":357,"key":"bigcodebench/current","label":"BigCodeBench","capability":"coding","harness":"published","source_url":"https://benchlm.ai/benchmarks/bigcodebench","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":92,"key":"bug-hunt-bench/current","label":"Bug Hunt Bench","capability":"coding","harness":"published","source_url":"https://benchlm.ai/benchmarks/bug-hunt-bench","featured":false,"n_measured":4,"n_estimated":0,"listed":false,"dense":false},{"id":72,"key":"cursor-bench31/current","label":"CursorBench 3.1","capability":"coding","harness":"published","source_url":"https://benchlm.ai/benchmarks","featured":false,"n_measured":7,"n_estimated":161,"listed":true,"dense":false},{"id":73,"key":"cursor-bench32/current","label":"CursorBench 3.2","capability":"coding","harness":"published","source_url":"https://benchlm.ai/benchmarks","featured":false,"n_measured":18,"n_estimated":160,"listed":true,"dense":true},{"id":93,"key":"cursor-bench40/current","label":"CursorBench 4.0","capability":"coding","harness":"published","source_url":"https://benchlm.ai/benchmarks","featured":false,"n_measured":13,"n_estimated":155,"listed":true,"dense":true},{"id":95,"key":"deepswe/current","label":"DeepSWE","capability":"coding","harness":"published","source_url":"https://benchlm.ai/benchmarks/deepswe","featured":false,"n_measured":43,"n_estimated":107,"listed":true,"dense":true},{"id":224,"key":"dsbenchfullstack/current","label":"DSBench-FullStack","capability":"coding","harness":"published","source_url":"https://benchlm.ai/benchmarks/dsbenchfullstack","featured":false,"n_measured":2,"n_estimated":0,"listed":false,"dense":false},{"id":225,"key":"dsbenchhard/current","label":"DSBench-Hard","capability":"coding","harness":"published","source_url":"https://benchlm.ai/benchmarks/dsbenchhard","featured":false,"n_measured":2,"n_estimated":0,"listed":false,"dense":false},{"id":275,"key":"eebench/current","label":"EEBench","capability":"coding","harness":"published","source_url":"https://benchlm.ai/benchmarks/eebench","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":75,"key":"frontiercode/current","label":"FrontierCode 1.1 Main","capability":"coding","harness":"published","source_url":"https://benchlm.ai/benchmarks/frontiercode","featured":false,"n_measured":16,"n_estimated":190,"listed":true,"dense":true},{"id":169,"key":"frontiercode11extended/current","label":"FrontierCode 1.1 Extended","capability":"coding","harness":"published","source_url":"https://benchlm.ai/benchmarks/frontiercode11extended","featured":false,"n_measured":8,"n_estimated":139,"listed":true,"dense":true},{"id":253,"key":"frontierswe/current","label":"FrontierSWE","capability":"coding","harness":"published","source_url":"https://benchlm.ai/benchmarks/frontierswe","featured":false,"n_measured":3,"n_estimated":0,"listed":false,"dense":false},{"id":76,"key":"frontierswev2/current","label":"FrontierSWE v2","capability":"coding","harness":"published","source_url":"https://benchlm.ai/benchmarks/frontierswev2","featured":false,"n_measured":18,"n_estimated":150,"listed":true,"dense":true},{"id":54,"key":"humaneval/current","label":"HumanEval","capability":"coding","harness":"published","source_url":"https://benchlm.ai/benchmarks/humaneval","featured":false,"n_measured":2,"n_estimated":0,"listed":false,"dense":false},{"id":327,"key":"kernelbenchhard/current","label":"KernelBench Hard","capability":"coding","harness":"published","source_url":"https://benchlm.ai/benchmarks/kernelbenchhard","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":301,"key":"kimicodebenchv2/current","label":"Kimi Code Bench v2","capability":"coding","harness":"published","source_url":"https://benchlm.ai/benchmarks/kimicodebenchv2","featured":false,"n_measured":2,"n_estimated":0,"listed":false,"dense":false},{"id":215,"key":"livecodebench/current","label":"LiveCodeBench","capability":"coding","harness":"published","source_url":"https://benchlm.ai/benchmarks/livecodebench","featured":false,"n_measured":7,"n_estimated":140,"listed":true,"dense":false},{"id":148,"key":"livecodebench-pro/current","label":"LiveCodeBench Pro","capability":"coding","harness":"published","source_url":"https://benchlm.ai/benchmarks/livecodebench-pro","featured":false,"n_measured":8,"n_estimated":67,"listed":true,"dense":true},{"id":319,"key":"livecodebench-v5/current","label":"LiveCodeBench v5","capability":"coding","harness":"published","source_url":"https://benchlm.ai/benchmarks/livecodebench-v5","featured":false,"n_measured":2,"n_estimated":0,"listed":false,"dense":false},{"id":17,"key":"livecodebench-v6/current","label":"LiveCodeBench v6","capability":"coding","harness":"published","source_url":"https://benchlm.ai/benchmarks/livecodebench-v6","featured":false,"n_measured":31,"n_estimated":157,"listed":true,"dense":true},{"id":227,"key":"livecodebenchpass1cot/current","label":"LiveCodeBench Pass@1-COT","capability":"coding","harness":"published","source_url":"https://benchlm.ai/benchmarks/livecodebenchpass1cot","featured":false,"n_measured":2,"n_estimated":0,"listed":false,"dense":false},{"id":303,"key":"mlsbenchlite/current","label":"MLS-Bench Lite","capability":"coding","harness":"published","source_url":"https://benchlm.ai/benchmarks/mlsbenchlite","featured":false,"n_measured":4,"n_estimated":0,"listed":false,"dense":false},{"id":324,"key":"multiswebench/current","label":"Multi-SWE Bench","capability":"coding","harness":"published","source_url":"https://benchlm.ai/benchmarks/multiswebench","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":135,"key":"nl2repo/current","label":"NL2Repo","capability":"coding","harness":"published","source_url":"https://benchlm.ai/benchmarks/nl2repo","featured":false,"n_measured":29,"n_estimated":168,"listed":true,"dense":true},{"id":218,"key":"openharmonybench/current","label":"OpenHarmony Bench","capability":"coding","harness":"published","source_url":"https://benchlm.ai/benchmarks/openharmonybench","featured":false,"n_measured":13,"n_estimated":157,"listed":true,"dense":true},{"id":351,"key":"paperbench/current","label":"PaperBench","capability":"coding","harness":"published","source_url":"https://benchlm.ai/benchmarks/paperbench","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":254,"key":"posttrainbench/current","label":"PostTrain Bench","capability":"coding","harness":"published","source_url":"https://benchlm.ai/benchmarks/posttrainbench","featured":false,"n_measured":3,"n_estimated":0,"listed":false,"dense":false},{"id":99,"key":"posttrainbench-v1-1/current","label":"PostTrainBench v1.1","capability":"coding","harness":"published","source_url":"https://benchlm.ai/benchmarks/posttrainbench-v1-1","featured":false,"n_measured":14,"n_estimated":175,"listed":true,"dense":true},{"id":100,"key":"programbench/current","label":"ProgramBench","capability":"coding","harness":"published","source_url":"https://benchlm.ai/benchmarks/programbench","featured":false,"n_measured":13,"n_estimated":18,"listed":true,"dense":true},{"id":188,"key":"programbenchepisode1/current","label":"ProgramBench (episode 1)","capability":"coding","harness":"published","source_url":"https://benchlm.ai/benchmarks/programbenchepisode1","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":152,"key":"reactnativeevals/current","label":"React Native Evals","capability":"coding","harness":"published","source_url":"https://benchlm.ai/benchmarks/reactnativeevals","featured":false,"n_measured":16,"n_estimated":151,"listed":true,"dense":true},{"id":19,"key":"scicode/current","label":"SciCode","capability":"coding","harness":"published","source_url":"https://benchlm.ai/benchmarks/scicode","featured":false,"n_measured":28,"n_estimated":168,"listed":true,"dense":true},{"id":282,"key":"spider2lite/current","label":"Spider 2.0-Lite","capability":"coding","harness":"published","source_url":"https://benchlm.ai/benchmarks/spider2lite","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":329,"key":"svgbench/current","label":"SVG-Bench","capability":"coding","harness":"published","source_url":"https://benchlm.ai/benchmarks/svgbench","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":320,"key":"swe-atlas-codebase-qna/current","label":"SWE-Atlas Codebase QnA","capability":"coding","harness":"published","source_url":"https://benchlm.ai/benchmarks/swe-atlas-codebase-qna","featured":false,"n_measured":3,"n_estimated":0,"listed":false,"dense":false},{"id":43,"key":"swe-bench-pro/current","label":"SWE-bench Pro","capability":"coding","harness":"published","source_url":"https://benchlm.ai/benchmarks/swe-bench-pro","featured":false,"n_measured":73,"n_estimated":129,"listed":true,"dense":true},{"id":37,"key":"swe-bench-verified/current","label":"SWE-bench Verified","capability":"coding","harness":"published","source_url":"https://benchlm.ai/benchmarks/swe-bench-verified","featured":false,"n_measured":77,"n_estimated":140,"listed":true,"dense":true},{"id":154,"key":"swe-rebench/current","label":"SWE-Rebench","capability":"coding","harness":"published","source_url":"https://benchlm.ai/benchmarks/swe-rebench","featured":false,"n_measured":13,"n_estimated":158,"listed":true,"dense":true},{"id":155,"key":"swe-verified-arcee/current","label":"SWE-bench Verified*","capability":"coding","harness":"published","source_url":"https://benchlm.ai/benchmarks/swe-bench-verified","featured":false,"n_measured":5,"n_estimated":0,"listed":false,"dense":false},{"id":255,"key":"swemarathon/current","label":"SWE-Marathon","capability":"coding","harness":"published","source_url":"https://benchlm.ai/benchmarks/swemarathon","featured":false,"n_measured":4,"n_estimated":0,"listed":false,"dense":false},{"id":81,"key":"valslivecodebench/current","label":"Vals LiveCodeBench","capability":"coding","harness":"vals-ai","source_url":"https://benchlm.ai/benchmarks/valslivecodebench","featured":false,"n_measured":58,"n_estimated":94,"listed":true,"dense":true},{"id":83,"key":"valsswebench/current","label":"Vals SWE-bench","capability":"coding","harness":"vals-ai","source_url":"https://benchlm.ai/benchmarks/valsswebench","featured":false,"n_measured":60,"n_estimated":150,"listed":true,"dense":true},{"id":108,"key":"vibecodebench/current","label":"Vibe Code Bench","capability":"coding","harness":"published","source_url":"https://benchlm.ai/benchmarks/vibecodebench","featured":false,"n_measured":43,"n_estimated":135,"listed":true,"dense":true},{"id":325,"key":"vibepro/current","label":"VIBE-Pro","capability":"coding","harness":"published","source_url":"https://benchlm.ai/benchmarks/vibepro","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":330,"key":"vibev2/current","label":"VIBE V2","capability":"coding","harness":"published","source_url":"https://benchlm.ai/benchmarks/vibev2","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":85,"key":"vulcanbench/current","label":"VulcanBench v3","capability":"coding","harness":"published","source_url":"https://benchlm.ai/benchmarks/vulcanbench","featured":false,"n_measured":14,"n_estimated":52,"listed":true,"dense":true},{"id":197,"key":"vulcanciiv1/current","label":"VulcanBench CII v1","capability":"coding","harness":"published","source_url":"https://benchlm.ai/benchmarks/vulcanciiv1","featured":false,"n_measured":3,"n_estimated":0,"listed":false,"dense":false},{"id":257,"key":"aa-aime2025/current","label":"AA AIME 2025","capability":"math","harness":"artificial-analysis","source_url":"https://benchlm.ai/benchmarks/aaaime2025","featured":true,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":271,"key":"aa-math500/current","label":"AA MATH-500","capability":"math","harness":"artificial-analysis","source_url":"https://benchlm.ai/benchmarks/math-500","featured":true,"n_measured":3,"n_estimated":0,"listed":false,"dense":false},{"id":338,"key":"aime2024/current","label":"AIME 2024","capability":"math","harness":"published","source_url":"https://benchlm.ai/benchmarks/aime2024","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":205,"key":"aime2025/current","label":"AIME 2025","capability":"math","harness":"published","source_url":"https://benchlm.ai/benchmarks/aime2025","featured":false,"n_measured":16,"n_estimated":23,"listed":true,"dense":true},{"id":145,"key":"aime2025arcee/current","label":"AIME25 (Arcee)","capability":"math","harness":"published","source_url":"https://benchlm.ai/benchmarks/aime2025arcee","featured":false,"n_measured":6,"n_estimated":0,"listed":false,"dense":false},{"id":4,"key":"aime2026/current","label":"AIME26","capability":"math","harness":"published","source_url":"https://benchlm.ai/benchmarks/aime2026","featured":false,"n_measured":29,"n_estimated":58,"listed":true,"dense":true},{"id":5,"key":"apex/current","label":"Apex","capability":"math","harness":"published","source_url":"https://benchlm.ai/benchmarks/apex","featured":false,"n_measured":8,"n_estimated":0,"listed":false,"dense":false},{"id":6,"key":"apexshortlist/current","label":"Apex Shortlist","capability":"math","harness":"published","source_url":"https://benchlm.ai/benchmarks/apexshortlist","featured":false,"n_measured":3,"n_estimated":0,"listed":false,"dense":false},{"id":198,"key":"arxivmathaugust2026/current","label":"ArXivMath Aug. 2026 (no tools)","capability":"math","harness":"published","source_url":"https://benchlm.ai/benchmarks/arxivmathaugust2026","featured":false,"n_measured":2,"n_estimated":0,"listed":false,"dense":false},{"id":199,"key":"arxivmathaugust2026withtools/current","label":"ArXivMath Aug. 2026 (tools)","capability":"math","harness":"published","source_url":"https://benchlm.ai/benchmarks/arxivmathaugust2026withtools","featured":false,"n_measured":2,"n_estimated":0,"listed":false,"dense":false},{"id":162,"key":"arxivmathjune2026/current","label":"ArXivMath Jun. 2026 (no tools)","capability":"math","harness":"published","source_url":"https://benchlm.ai/benchmarks/arxivmathjune2026","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":163,"key":"arxivmathjune2026withtools/current","label":"ArXivMath Jun. 2026 (tools)","capability":"math","harness":"published","source_url":"https://benchlm.ai/benchmarks/arxivmathjune2026withtools","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":158,"key":"frontiermath/current","label":"FrontierMath (legacy)","capability":"math","harness":"published","source_url":"https://benchlm.ai/benchmarks/frontiermath","featured":false,"n_measured":7,"n_estimated":46,"listed":true,"dense":false},{"id":59,"key":"frontiermathv2tier4/current","label":"FrontierMath v2 (Tier 4)","capability":"math","harness":"published","source_url":"https://benchlm.ai/benchmarks/frontiermathv2tier4","featured":false,"n_measured":45,"n_estimated":12,"listed":true,"dense":true},{"id":60,"key":"frontiermathv2tiers13/current","label":"FrontierMath v2 (Tiers 1-3)","capability":"math","harness":"published","source_url":"https://benchlm.ai/benchmarks/frontiermathv2tiers13","featured":false,"n_measured":51,"n_estimated":19,"listed":true,"dense":true},{"id":57,"key":"gsm8k/current","label":"GSM8K","capability":"math","harness":"published","source_url":"https://benchlm.ai/benchmarks/gsm8k","featured":false,"n_measured":3,"n_estimated":0,"listed":false,"dense":false},{"id":125,"key":"hmmtfeb2025/current","label":"HMMT Feb 2025","capability":"math","harness":"published","source_url":"https://benchlm.ai/benchmarks/hmmtfeb2025","featured":false,"n_measured":10,"n_estimated":26,"listed":true,"dense":true},{"id":126,"key":"hmmtfeb2026/current","label":"HMMT Feb 2026","capability":"math","harness":"published","source_url":"https://benchlm.ai/benchmarks/hmmtfeb2026","featured":false,"n_measured":23,"n_estimated":14,"listed":true,"dense":true},{"id":127,"key":"hmmtnov2025/current","label":"HMMT Nov 2025","capability":"math","harness":"published","source_url":"https://benchlm.ai/benchmarks/hmmtnov2025","featured":false,"n_measured":9,"n_estimated":43,"listed":true,"dense":true},{"id":36,"key":"imo2026/current","label":"IMO 2026","capability":"math","harness":"published","source_url":"https://benchlm.ai/benchmarks/imo2026","featured":false,"n_measured":2,"n_estimated":0,"listed":false,"dense":false},{"id":226,"key":"imoanswerbench/current","label":"IMOAnswerBench","capability":"math","harness":"published","source_url":"https://benchlm.ai/benchmarks/imoanswerbench","featured":false,"n_measured":9,"n_estimated":15,"listed":true,"dense":true},{"id":260,"key":"ipho2025theory/current","label":"IPhO 2025 (Theory)","capability":"math","harness":"published","source_url":"https://benchlm.ai/benchmarks/ipho2025theory","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":298,"key":"math-500/current","label":"MATH-500","capability":"math","harness":"published","source_url":"https://benchlm.ai/benchmarks/math-500","featured":false,"n_measured":7,"n_estimated":24,"listed":true,"dense":false},{"id":131,"key":"mmanswerbench/current","label":"MMAnswerBench","capability":"math","harness":"published","source_url":"https://benchlm.ai/benchmarks/mmanswerbench","featured":false,"n_measured":10,"n_estimated":72,"listed":true,"dense":true},{"id":193,"key":"riemannbench/current","label":"RiemannBench (no tools)","capability":"math","harness":"published","source_url":"https://benchlm.ai/benchmarks/riemannbench","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":194,"key":"riemannbenchwithtools/current","label":"RiemannBench (tools)","capability":"math","harness":"published","source_url":"https://benchlm.ai/benchmarks/riemannbenchwithtools","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":117,"key":"usamo2026/current","label":"USAMO 2026","capability":"math","harness":"published","source_url":"https://benchlm.ai/benchmarks/usamo2026","featured":false,"n_measured":3,"n_estimated":0,"listed":false,"dense":false},{"id":1,"key":"aa-gpqa-diamond/current","label":"AA-GPQA Diamond","capability":"knowledge","harness":"artificial-analysis","source_url":"https://benchlm.ai/benchmarks/gpqa-diamond","featured":true,"n_measured":181,"n_estimated":63,"listed":true,"dense":true},{"id":2,"key":"aa-hle/current","label":"AA-HLE","capability":"knowledge","harness":"artificial-analysis","source_url":"https://benchlm.ai/benchmarks/aahle","featured":true,"n_measured":195,"n_estimated":47,"listed":true,"dense":true},{"id":120,"key":"aa-mmlu-pro/current","label":"AA MMLU-Pro","capability":"knowledge","harness":"artificial-analysis","source_url":"https://benchlm.ai/benchmarks/mmlu-pro","featured":true,"n_measured":3,"n_estimated":0,"listed":false,"dense":false},{"id":354,"key":"agieval/current","label":"AGIEval","capability":"knowledge","harness":"published","source_url":"https://benchlm.ai/benchmarks/agieval","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":121,"key":"aineedle/current","label":"AI-Needle","capability":"knowledge","harness":"published","source_url":"https://benchlm.ai/benchmarks/aineedle","featured":false,"n_measured":4,"n_estimated":0,"listed":false,"dense":false},{"id":70,"key":"arc-agi-2/current","label":"ARC-AGI-2","capability":"knowledge","harness":"published","source_url":"https://benchlm.ai/benchmarks/arc-agi-2","featured":false,"n_measured":41,"n_estimated":200,"listed":true,"dense":true},{"id":69,"key":"arcagi1/current","label":"ARC-AGI-1","capability":"knowledge","harness":"published","source_url":"https://benchlm.ai/benchmarks/arcagi1","featured":false,"n_measured":35,"n_estimated":207,"listed":true,"dense":true},{"id":157,"key":"arcagi3/current","label":"ARC-AGI-3","capability":"knowledge","harness":"published","source_url":"https://benchlm.ai/benchmarks/arcagi3","featured":false,"n_measured":17,"n_estimated":222,"listed":true,"dense":true},{"id":7,"key":"artificialanalysis/current","label":"Artificial Analysis Intelligence Index","capability":"knowledge","harness":"artificial-analysis","source_url":"https://benchlm.ai/benchmarks/artificialanalysis","featured":false,"n_measured":199,"n_estimated":45,"listed":true,"dense":true},{"id":251,"key":"bbh/current","label":"BBH","capability":"knowledge","harness":"published","source_url":"https://benchlm.ai/benchmarks/bbh","featured":false,"n_measured":3,"n_estimated":0,"listed":false,"dense":false},{"id":34,"key":"biomysterybenchhumandifficult/current","label":"BioMysteryBench (human-difficult)","capability":"knowledge","harness":"published","source_url":"https://benchlm.ai/benchmarks/biomysterybenchhumandifficult","featured":false,"n_measured":6,"n_estimated":197,"listed":true,"dense":false},{"id":166,"key":"biomysterybenchhumansolvable/current","label":"BioMysteryBench (human-solvable)","capability":"knowledge","harness":"published","source_url":"https://benchlm.ai/benchmarks/biomysterybenchhumansolvable","featured":false,"n_measured":5,"n_estimated":198,"listed":true,"dense":false},{"id":122,"key":"ceval/current","label":"C-Eval","capability":"knowledge","harness":"published","source_url":"https://benchlm.ai/benchmarks/ceval","featured":false,"n_measured":6,"n_estimated":225,"listed":true,"dense":false},{"id":222,"key":"chinesesimpleqa/current","label":"Chinese-SimpleQA","capability":"knowledge","harness":"published","source_url":"https://benchlm.ai/benchmarks/chinesesimpleqa","featured":false,"n_measured":2,"n_estimated":0,"listed":false,"dense":false},{"id":321,"key":"cmmlu/current","label":"CMMLU","capability":"knowledge","harness":"published","source_url":"https://benchlm.ai/benchmarks/cmmlu","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":9,"key":"critpt/current","label":"CritPt","capability":"knowledge","harness":"artificial-analysis","source_url":"https://benchlm.ai/benchmarks/critpt","featured":true,"n_measured":190,"n_estimated":52,"listed":true,"dense":true},{"id":56,"key":"drop/current","label":"DROP","capability":"knowledge","harness":"published","source_url":"https://benchlm.ai/benchmarks/drop","featured":false,"n_measured":2,"n_estimated":0,"listed":false,"dense":false},{"id":259,"key":"frontierscience/current","label":"FrontierScience","capability":"knowledge","harness":"published","source_url":"https://benchlm.ai/benchmarks/frontierscience","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":26,"key":"frontierscienceresearch/current","label":"FrontierScience Research","capability":"knowledge","harness":"published","source_url":"https://benchlm.ai/benchmarks/frontierscienceresearch","featured":false,"n_measured":4,"n_estimated":0,"listed":false,"dense":false},{"id":270,"key":"genebenchpro/current","label":"GeneBench-Pro","capability":"knowledge","harness":"published","source_url":"https://benchlm.ai/benchmarks/genebenchpro","featured":false,"n_measured":2,"n_estimated":0,"listed":false,"dense":false},{"id":61,"key":"gpqa/current","label":"GPQA","capability":"knowledge","harness":"published","source_url":"https://benchlm.ai/benchmarks/gpqa","featured":false,"n_measured":84,"n_estimated":160,"listed":true,"dense":true},{"id":11,"key":"gpqa-diamond/current","label":"GPQA Diamond","capability":"knowledge","harness":"published","source_url":"https://benchlm.ai/benchmarks/gpqa-diamond","featured":false,"n_measured":67,"n_estimated":177,"listed":true,"dense":true},{"id":96,"key":"graphwalks-bfs-256k-1m/current","label":"GraphWalks BFS 256K–1M","capability":"knowledge","harness":"published","source_url":"https://benchlm.ai/benchmarks/graphwalks-bfs-256k-1m","featured":false,"n_measured":4,"n_estimated":0,"listed":false,"dense":false},{"id":173,"key":"healthbench/current","label":"HealthBench (raw)","capability":"knowledge","harness":"published","source_url":"https://benchlm.ai/benchmarks/healthbench","featured":false,"n_measured":7,"n_estimated":192,"listed":true,"dense":false},{"id":147,"key":"healthbench-hard/current","label":"HealthBench Hard","capability":"knowledge","harness":"published","source_url":"https://benchlm.ai/benchmarks/healthbench-hard","featured":false,"n_measured":12,"n_estimated":213,"listed":true,"dense":true},{"id":174,"key":"healthbenchlengthadjusted/current","label":"HealthBench (length-adjusted)","capability":"knowledge","harness":"published","source_url":"https://benchlm.ai/benchmarks/healthbenchlengthadjusted","featured":false,"n_measured":7,"n_estimated":196,"listed":true,"dense":false},{"id":175,"key":"healthbenchprofessional/current","label":"HealthBench Professional","capability":"knowledge","harness":"published","source_url":"https://benchlm.ai/benchmarks/healthbenchprofessional","featured":false,"n_measured":14,"n_estimated":226,"listed":true,"dense":true},{"id":176,"key":"healthbenchprofessionalraw/current","label":"HealthBench Professional (raw)","capability":"knowledge","harness":"published","source_url":"https://benchlm.ai/benchmarks/healthbenchprofessionalraw","featured":false,"n_measured":7,"n_estimated":193,"listed":true,"dense":false},{"id":12,"key":"hle/current","label":"HLE","capability":"knowledge","harness":"published","source_url":"https://benchlm.ai/benchmarks/hle","featured":false,"n_measured":59,"n_estimated":185,"listed":true,"dense":true},{"id":97,"key":"hlenotools/current","label":"HLE w/o tools","capability":"knowledge","harness":"published","source_url":"https://benchlm.ai/benchmarks/hlenotools","featured":false,"n_measured":42,"n_estimated":200,"listed":true,"dense":true},{"id":177,"key":"hleverified/current","label":"HLE-Verified","capability":"knowledge","harness":"published","source_url":"https://benchlm.ai/benchmarks/hleverified","featured":false,"n_measured":6,"n_estimated":206,"listed":true,"dense":false},{"id":47,"key":"jev-bench14/current","label":"JevBench 1.4","capability":"knowledge","harness":"published","source_url":"https://benchlm.ai/benchmarks","featured":false,"n_measured":93,"n_estimated":18,"listed":true,"dense":false},{"id":46,"key":"jev-bench15/current","label":"JevBench 1.5","capability":"knowledge","harness":"published","source_url":"https://benchlm.ai/benchmarks","featured":false,"n_measured":108,"n_estimated":0,"listed":false,"dense":false},{"id":178,"key":"labbench2/current","label":"LABBench2","capability":"knowledge","harness":"published","source_url":"https://benchlm.ai/benchmarks/labbench2","featured":false,"n_measured":7,"n_estimated":205,"listed":true,"dense":false},{"id":201,"key":"medicinalchemistry/current","label":"Medicinal chemistry","capability":"knowledge","harness":"published","source_url":"https://benchlm.ai/benchmarks/medicinalchemistry","featured":false,"n_measured":2,"n_estimated":0,"listed":false,"dense":false},{"id":150,"key":"medxpertqatext/current","label":"MedXpertQA (Text)","capability":"knowledge","harness":"published","source_url":"https://benchlm.ai/benchmarks/medxpertqatext","featured":false,"n_measured":5,"n_estimated":78,"listed":true,"dense":false},{"id":256,"key":"mmlu/current","label":"MMLU","capability":"knowledge","harness":"published","source_url":"https://benchlm.ai/benchmarks/mmlu","featured":false,"n_measured":7,"n_estimated":215,"listed":true,"dense":false},{"id":58,"key":"mmlu-pro/current","label":"MMLU-Pro","capability":"knowledge","harness":"published","source_url":"https://benchlm.ai/benchmarks/mmlu-pro","featured":false,"n_measured":46,"n_estimated":190,"listed":true,"dense":true},{"id":151,"key":"mmlu-pro-arcee/current","label":"MMLU-Pro (Arcee)","capability":"knowledge","harness":"published","source_url":"https://benchlm.ai/benchmarks/mmlu-pro","featured":false,"n_measured":6,"n_estimated":220,"listed":true,"dense":false},{"id":133,"key":"mmluredux/current","label":"MMLU-Redux","capability":"knowledge","harness":"published","source_url":"https://benchlm.ai/benchmarks/mmluredux","featured":false,"n_measured":11,"n_estimated":223,"listed":true,"dense":true},{"id":252,"key":"mmmlu/current","label":"MMMLU","capability":"knowledge","harness":"published","source_url":"https://benchlm.ai/benchmarks/mmmlu","featured":false,"n_measured":5,"n_estimated":62,"listed":true,"dense":false},{"id":202,"key":"morphologytomolecule/current","label":"Morphology-to-molecule matching","capability":"knowledge","harness":"published","source_url":"https://benchlm.ai/benchmarks/morphologytomolecule","featured":false,"n_measured":2,"n_estimated":0,"listed":false,"dense":false},{"id":18,"key":"omniscienceaccuracy/current","label":"AA-Omniscience Accuracy","capability":"knowledge","harness":"artificial-analysis","source_url":"https://benchlm.ai/benchmarks/omniscienceaccuracy","featured":false,"n_measured":183,"n_estimated":61,"listed":true,"dense":true},{"id":187,"key":"organicchemistryv2/current","label":"Organic chemistry V2","capability":"knowledge","harness":"published","source_url":"https://benchlm.ai/benchmarks/organicchemistryv2","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":203,"key":"proteinbinderdesign/current","label":"De novo protein-binder design","capability":"knowledge","harness":"published","source_url":"https://benchlm.ai/benchmarks/proteinbinderdesign","featured":false,"n_measured":2,"n_estimated":0,"listed":false,"dense":false},{"id":189,"key":"proteindesign/current","label":"Protein Design","capability":"knowledge","harness":"published","source_url":"https://benchlm.ai/benchmarks/proteindesign","featured":false,"n_measured":3,"n_estimated":0,"listed":false,"dense":false},{"id":204,"key":"proteindesignlibraryranking/current","label":"Protein Design library ranking","capability":"knowledge","harness":"published","source_url":"https://benchlm.ai/benchmarks/proteindesignlibraryranking","featured":false,"n_measured":2,"n_estimated":0,"listed":false,"dense":false},{"id":190,"key":"proteingymhard/current","label":"ProteinGym Hard","capability":"knowledge","harness":"published","source_url":"https://benchlm.ai/benchmarks/proteingymhard","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":191,"key":"protocolstroubleshooting/current","label":"Protocols (troubleshooting)","capability":"knowledge","harness":"published","source_url":"https://benchlm.ai/benchmarks/protocolstroubleshooting","featured":false,"n_measured":3,"n_estimated":0,"listed":false,"dense":false},{"id":192,"key":"protocolsunderstanding/current","label":"Protocols (understanding)","capability":"knowledge","harness":"published","source_url":"https://benchlm.ai/benchmarks/protocolsunderstanding","featured":false,"n_measured":3,"n_estimated":0,"listed":false,"dense":false},{"id":229,"key":"simpleqa/current","label":"SimpleQA","capability":"knowledge","harness":"published","source_url":"https://benchlm.ai/benchmarks/simpleqa","featured":false,"n_measured":3,"n_estimated":0,"listed":false,"dense":false},{"id":195,"key":"singlecellbench/current","label":"SingleCellBench","capability":"knowledge","harness":"published","source_url":"https://benchlm.ai/benchmarks/singlecellbench","featured":false,"n_measured":3,"n_estimated":0,"listed":false,"dense":false},{"id":196,"key":"spatialbenchverified/current","label":"SpatialBench Verified","capability":"knowledge","harness":"published","source_url":"https://benchlm.ai/benchmarks/spatialbenchverified","featured":false,"n_measured":3,"n_estimated":0,"listed":false,"dense":false},{"id":140,"key":"supergpqa/current","label":"SuperGPQA","capability":"knowledge","harness":"published","source_url":"https://benchlm.ai/benchmarks/supergpqa","featured":false,"n_measured":20,"n_estimated":214,"listed":true,"dense":true},{"id":80,"key":"vals-gpqa-diamond/current","label":"Vals GPQA Diamond","capability":"knowledge","harness":"vals-ai","source_url":"https://benchlm.ai/benchmarks/gpqa-diamond","featured":false,"n_measured":56,"n_estimated":185,"listed":true,"dense":true},{"id":82,"key":"vals-mmlu-pro/current","label":"Vals MMLU-Pro","capability":"knowledge","harness":"vals-ai","source_url":"https://benchlm.ai/benchmarks/mmlu-pro","featured":false,"n_measured":56,"n_estimated":186,"listed":true,"dense":true},{"id":62,"key":"aa-if-bench/current","label":"AA-IFBench","capability":"instruction-following","harness":"artificial-analysis","source_url":"https://benchlm.ai/benchmarks/ifbench","featured":true,"n_measured":130,"n_estimated":38,"listed":true,"dense":true},{"id":13,"key":"ifbench/current","label":"IFBench","capability":"instruction-following","harness":"published","source_url":"https://benchlm.ai/benchmarks/ifbench","featured":false,"n_measured":43,"n_estimated":125,"listed":true,"dense":true},{"id":23,"key":"ifeval/current","label":"IFEval","capability":"instruction-following","harness":"published","source_url":"https://benchlm.ai/benchmarks/ifeval","featured":false,"n_measured":33,"n_estimated":135,"listed":true,"dense":true},{"id":281,"key":"sobvalueacc/current","label":"SOB Value Acc","capability":"instruction-following","harness":"published","source_url":"https://benchlm.ai/benchmarks/sobvalueacc","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":156,"key":"aa-global-mmlu-lite/current","label":"AA Global-MMLU-Lite","capability":"multilingual","harness":"artificial-analysis","source_url":"https://benchlm.ai/benchmarks/aaglobalmmlulite","featured":true,"n_measured":4,"n_estimated":0,"listed":false,"dense":false},{"id":8,"key":"click/current","label":"CLIcK","capability":"multilingual","harness":"published","source_url":"https://benchlm.ai/benchmarks/click","featured":false,"n_measured":3,"n_estimated":0,"listed":false,"dense":false},{"id":172,"key":"gmmlu/current","label":"GMMLU","capability":"multilingual","harness":"published","source_url":"https://benchlm.ai/benchmarks/gmmlu","featured":false,"n_measured":3,"n_estimated":0,"listed":false,"dense":false},{"id":353,"key":"hrm8k/current","label":"HRM8K","capability":"multilingual","harness":"published","source_url":"https://benchlm.ai/benchmarks/hrm8k","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":161,"key":"include/current","label":"INCLUDE","capability":"multilingual","harness":"published","source_url":"https://benchlm.ai/benchmarks/include","featured":false,"n_measured":4,"n_estimated":0,"listed":false,"dense":false},{"id":297,"key":"kmmlu/current","label":"KMMLU","capability":"multilingual","harness":"published","source_url":"https://benchlm.ai/benchmarks/kmmlu","featured":false,"n_measured":2,"n_estimated":0,"listed":false,"dense":false},{"id":14,"key":"kmmlupro/current","label":"KMMLU-Pro","capability":"multilingual","harness":"published","source_url":"https://benchlm.ai/benchmarks/kmmlupro","featured":false,"n_measured":4,"n_estimated":0,"listed":false,"dense":false},{"id":15,"key":"kobalt/current","label":"KoBALT","capability":"multilingual","harness":"published","source_url":"https://benchlm.ai/benchmarks/kobalt","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":346,"key":"maxife/current","label":"MAXIFE","capability":"multilingual","harness":"published","source_url":"https://benchlm.ai/benchmarks/maxife","featured":false,"n_measured":2,"n_estimated":0,"listed":false,"dense":false},{"id":184,"key":"milu/current","label":"MILU","capability":"multilingual","harness":"published","source_url":"https://benchlm.ai/benchmarks/milu","featured":false,"n_measured":3,"n_estimated":0,"listed":false,"dense":false},{"id":312,"key":"mkqa11/current","label":"MKQA-11","capability":"multilingual","harness":"published","source_url":"https://benchlm.ai/benchmarks/mkqa11","featured":false,"n_measured":2,"n_estimated":0,"listed":false,"dense":false},{"id":132,"key":"mmluprox/current","label":"MMLU-ProX","capability":"multilingual","harness":"published","source_url":"https://benchlm.ai/benchmarks/mmluprox","featured":false,"n_measured":16,"n_estimated":38,"listed":true,"dense":true},{"id":313,"key":"nanobeirmultilingual/current","label":"NanoBEIR Multilingual","capability":"multilingual","harness":"published","source_url":"https://benchlm.ai/benchmarks/nanobeirmultilingual","featured":false,"n_measured":2,"n_estimated":0,"listed":false,"dense":false},{"id":136,"key":"nova63/current","label":"NOVA-63","capability":"multilingual","harness":"published","source_url":"https://benchlm.ai/benchmarks/nova63","featured":false,"n_measured":7,"n_estimated":47,"listed":true,"dense":false},{"id":296,"key":"polymath/current","label":"PolyMath","capability":"multilingual","harness":"published","source_url":"https://benchlm.ai/benchmarks/polymath","featured":false,"n_measured":3,"n_estimated":0,"listed":false,"dense":false},{"id":101,"key":"swe-bench-multilingual/current","label":"SWE Multilingual","capability":"multilingual","harness":"published","source_url":"https://benchlm.ai/benchmarks/swe-bench-multilingual","featured":false,"n_measured":45,"n_estimated":9,"listed":true,"dense":true},{"id":337,"key":"wmt26/current","label":"WMT26","capability":"multilingual","harness":"published","source_url":"https://benchlm.ai/benchmarks/wmt26","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":31,"key":"aa-mmmu-pro/current","label":"AA-MMMU-Pro","capability":"vision","harness":"artificial-analysis","source_url":"https://benchlm.ai/benchmarks/mmmu-pro","featured":true,"n_measured":104,"n_estimated":25,"listed":true,"dense":true},{"id":333,"key":"ai2dtest/current","label":"AI2D_TEST","capability":"vision","harness":"published","source_url":"https://benchlm.ai/benchmarks/ai2dtest","featured":false,"n_measured":3,"n_estimated":0,"listed":false,"dense":false},{"id":355,"key":"aokvqa/current","label":"A-OKVQA","capability":"vision","harness":"published","source_url":"https://benchlm.ai/benchmarks/aokvqa","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":230,"key":"babyvision/current","label":"BabyVision","capability":"vision","harness":"published","source_url":"https://benchlm.ai/benchmarks/babyvision","featured":false,"n_measured":7,"n_estimated":14,"listed":true,"dense":false},{"id":217,"key":"babyvisionpython/current","label":"BabyVision w/ Python","capability":"vision","harness":"published","source_url":"https://benchlm.ai/benchmarks/babyvisionpython","featured":false,"n_measured":4,"n_estimated":0,"listed":false,"dense":false},{"id":164,"key":"benchcadvision2code/current","label":"BenchCAD Vision2Code (no tools)","capability":"vision","harness":"published","source_url":"https://benchlm.ai/benchmarks/benchcadvision2code","featured":false,"n_measured":3,"n_estimated":0,"listed":false,"dense":false},{"id":165,"key":"benchcadvision2codewithtools/current","label":"BenchCAD Vision2Code (tools)","capability":"vision","harness":"published","source_url":"https://benchlm.ai/benchmarks/benchcadvision2codewithtools","featured":false,"n_measured":4,"n_estimated":0,"listed":false,"dense":false},{"id":200,"key":"biomedicalimageanalysis/current","label":"Biomedical image analysis","capability":"vision","harness":"published","source_url":"https://benchlm.ai/benchmarks/biomedicalimageanalysis","featured":false,"n_measured":2,"n_estimated":0,"listed":false,"dense":false},{"id":71,"key":"blueprintbench2/current","label":"Blueprint-Bench 2","capability":"vision","harness":"published","source_url":"https://benchlm.ai/benchmarks/blueprintbench2","featured":false,"n_measured":2,"n_estimated":0,"listed":false,"dense":false},{"id":341,"key":"ccocr/current","label":"CC-OCR","capability":"vision","harness":"published","source_url":"https://benchlm.ai/benchmarks/ccocr","featured":false,"n_measured":3,"n_estimated":0,"listed":false,"dense":false},{"id":109,"key":"chartography/current","label":"Chartography (no tools)","capability":"vision","harness":"published","source_url":"https://benchlm.ai/benchmarks/chartography","featured":false,"n_measured":5,"n_estimated":0,"listed":false,"dense":false},{"id":167,"key":"chartographywithtools/current","label":"Chartography (tools)","capability":"vision","harness":"published","source_url":"https://benchlm.ai/benchmarks/chartographywithtools","featured":false,"n_measured":5,"n_estimated":0,"listed":false,"dense":false},{"id":112,"key":"charxiv/current","label":"CharXiv","capability":"vision","harness":"published","source_url":"https://benchlm.ai/benchmarks/charxiv","featured":false,"n_measured":39,"n_estimated":91,"listed":true,"dense":true},{"id":113,"key":"charxivnotools/current","label":"CharXiv w/o tools","capability":"vision","harness":"published","source_url":"https://benchlm.ai/benchmarks/charxivnotools","featured":false,"n_measured":14,"n_estimated":108,"listed":true,"dense":true},{"id":358,"key":"charxivoverall/current","label":"CharXiv (overall)","capability":"vision","harness":"published","source_url":"https://benchlm.ai/benchmarks/charxivoverall","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":317,"key":"countbench/current","label":"CountBench","capability":"vision","harness":"published","source_url":"https://benchlm.ai/benchmarks/countbench","featured":false,"n_measured":4,"n_estimated":0,"listed":false,"dense":false},{"id":342,"key":"dynamath/current","label":"DynaMath","capability":"vision","harness":"published","source_url":"https://benchlm.ai/benchmarks/dynamath","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":146,"key":"erqa/current","label":"ERQA","capability":"vision","harness":"published","source_url":"https://benchlm.ai/benchmarks/erqa","featured":false,"n_measured":13,"n_estimated":113,"listed":true,"dense":true},{"id":170,"key":"gdppdf/current","label":"GDP.pdf (no tools)","capability":"vision","harness":"published","source_url":"https://benchlm.ai/benchmarks/gdppdf","featured":false,"n_measured":2,"n_estimated":0,"listed":false,"dense":false},{"id":171,"key":"gdppdfwithtools/current","label":"GDP.pdf (tools)","capability":"vision","harness":"published","source_url":"https://benchlm.ai/benchmarks/gdppdfwithtools","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":314,"key":"liquidextractjsonvalidity/current","label":"Liquid Extract JSON Validity","capability":"vision","harness":"published","source_url":"https://benchlm.ai/benchmarks/liquidextractjsonvalidity","featured":false,"n_measured":2,"n_estimated":0,"listed":false,"dense":false},{"id":315,"key":"liquidextractschemaf1/current","label":"Liquid Extract F1","capability":"vision","harness":"published","source_url":"https://benchlm.ai/benchmarks/liquidextractschemaf1","featured":false,"n_measured":2,"n_estimated":0,"listed":false,"dense":false},{"id":316,"key":"liquidextractvlmjudge/current","label":"Liquid Extract VLM Judge","capability":"vision","harness":"published","source_url":"https://benchlm.ai/benchmarks/liquidextractvlmjudge","featured":false,"n_measured":2,"n_estimated":0,"listed":false,"dense":false},{"id":245,"key":"lvbench/current","label":"LVBench","capability":"vision","harness":"published","source_url":"https://benchlm.ai/benchmarks/lvbench","featured":false,"n_measured":6,"n_estimated":0,"listed":false,"dense":false},{"id":128,"key":"mathvision/current","label":"MathVision","capability":"vision","harness":"published","source_url":"https://benchlm.ai/benchmarks/mathvision","featured":false,"n_measured":19,"n_estimated":107,"listed":true,"dense":true},{"id":307,"key":"mathvisionpython/current","label":"MathVision w/ Python","capability":"vision","harness":"published","source_url":"https://benchlm.ai/benchmarks/mathvisionpython","featured":false,"n_measured":5,"n_estimated":39,"listed":true,"dense":false},{"id":149,"key":"medxpertqamm/current","label":"MedXpertQA (MM)","capability":"vision","harness":"published","source_url":"https://benchlm.ai/benchmarks/medxpertqamm","featured":false,"n_measured":8,"n_estimated":114,"listed":true,"dense":true},{"id":344,"key":"mlvuavg/current","label":"MLVU (M-Avg)","capability":"vision","harness":"published","source_url":"https://benchlm.ai/benchmarks/mlvuavg","featured":false,"n_measured":4,"n_estimated":0,"listed":false,"dense":false},{"id":207,"key":"mmmu/current","label":"MMMU","capability":"vision","harness":"published","source_url":"https://benchlm.ai/benchmarks/mmmu","featured":false,"n_measured":11,"n_estimated":108,"listed":true,"dense":true},{"id":134,"key":"mmmu-pro/current","label":"MMMU-Pro","capability":"vision","harness":"published","source_url":"https://benchlm.ai/benchmarks/mmmu-pro","featured":false,"n_measured":42,"n_estimated":88,"listed":true,"dense":true},{"id":258,"key":"mmmupropython/current","label":"MMMU-Pro w/ Python","capability":"vision","harness":"published","source_url":"https://benchlm.ai/benchmarks/mmmupropython","featured":false,"n_measured":9,"n_estimated":105,"listed":true,"dense":true},{"id":347,"key":"mmsearchplus/current","label":"MMSearch-Plus","capability":"vision","harness":"published","source_url":"https://benchlm.ai/benchmarks/mmsearchplus","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":231,"key":"mmvu/current","label":"MMVU","capability":"vision","harness":"published","source_url":"https://benchlm.ai/benchmarks/mmvu","featured":false,"n_measured":7,"n_estimated":14,"listed":true,"dense":false},{"id":343,"key":"mstar/current","label":"MStar","capability":"vision","harness":"published","source_url":"https://benchlm.ai/benchmarks/mstar","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":278,"key":"ocrbenchv2/current","label":"OCRBench V2","capability":"vision","harness":"published","source_url":"https://benchlm.ai/benchmarks/ocrbenchv2","featured":false,"n_measured":5,"n_estimated":0,"listed":false,"dense":false},{"id":345,"key":"odinw13/current","label":"ODINW13","capability":"vision","harness":"published","source_url":"https://benchlm.ai/benchmarks/odinw13","featured":false,"n_measured":2,"n_estimated":0,"listed":false,"dense":false},{"id":186,"key":"officeqa/current","label":"OfficeQA","capability":"vision","harness":"published","source_url":"https://benchlm.ai/benchmarks/officeqa","featured":false,"n_measured":3,"n_estimated":0,"listed":false,"dense":false},{"id":77,"key":"officeqapro/current","label":"OfficeQA Pro","capability":"vision","harness":"published","source_url":"https://benchlm.ai/benchmarks/officeqapro","featured":false,"n_measured":13,"n_estimated":96,"listed":true,"dense":true},{"id":279,"key":"olmocr/current","label":"olmOCR","capability":"vision","harness":"published","source_url":"https://benchlm.ai/benchmarks/olmocr","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":308,"key":"omnidocbench/current","label":"OmniDocBench","capability":"vision","harness":"published","source_url":"https://benchlm.ai/benchmarks/omnidocbench","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":328,"key":"omnidocbench15/current","label":"OmniDocBench 1.5","capability":"vision","harness":"published","source_url":"https://benchlm.ai/benchmarks/omnidocbench15","featured":false,"n_measured":6,"n_estimated":116,"listed":true,"dense":false},{"id":359,"key":"omnidocbench16/current","label":"OmniDocBench 1.6","capability":"vision","harness":"published","source_url":"https://benchlm.ai/benchmarks/omnidocbench16","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":232,"key":"perceptionbench/current","label":"PerceptionBench","capability":"vision","harness":"published","source_url":"https://benchlm.ai/benchmarks/perceptionbench","featured":false,"n_measured":3,"n_estimated":0,"listed":false,"dense":false},{"id":318,"key":"realworldqa/current","label":"RealWorldQA","capability":"vision","harness":"published","source_url":"https://benchlm.ai/benchmarks/realworldqa","featured":false,"n_measured":12,"n_estimated":114,"listed":true,"dense":true},{"id":280,"key":"refcocoavg/current","label":"RefCOCO (avg)","capability":"vision","harness":"published","source_url":"https://benchlm.ai/benchmarks/refcocoavg","featured":false,"n_measured":6,"n_estimated":6,"listed":true,"dense":false},{"id":139,"key":"screenspot-pro/current","label":"ScreenSpot Pro","capability":"vision","harness":"published","source_url":"https://benchlm.ai/benchmarks/screenspot-pro","featured":false,"n_measured":18,"n_estimated":108,"listed":true,"dense":true},{"id":233,"key":"simplevqa/current","label":"SimpleVQA","capability":"vision","harness":"published","source_url":"https://benchlm.ai/benchmarks/simplevqa","featured":false,"n_measured":11,"n_estimated":57,"listed":true,"dense":true},{"id":102,"key":"swe-bench-multimodal/current","label":"SWE-bench Multimodal","capability":"vision","harness":"published","source_url":"https://benchlm.ai/benchmarks/swe-bench-multimodal","featured":false,"n_measured":7,"n_estimated":0,"listed":false,"dense":false},{"id":299,"key":"videomme/current","label":"Video-MME","capability":"vision","harness":"published","source_url":"https://benchlm.ai/benchmarks/videomme","featured":false,"n_measured":3,"n_estimated":0,"listed":false,"dense":false},{"id":335,"key":"videommenosub/current","label":"Video-MME (w/o subtitle)","capability":"vision","harness":"published","source_url":"https://benchlm.ai/benchmarks/videommenosub","featured":false,"n_measured":2,"n_estimated":0,"listed":false,"dense":false},{"id":323,"key":"videommewithsub/current","label":"Video-MME (with subtitle)","capability":"vision","harness":"published","source_url":"https://benchlm.ai/benchmarks/videommewithsub","featured":false,"n_measured":6,"n_estimated":52,"listed":true,"dense":false},{"id":144,"key":"videommmu/current","label":"VideoMMMU","capability":"vision","harness":"published","source_url":"https://benchlm.ai/benchmarks/videommmu","featured":false,"n_measured":11,"n_estimated":55,"listed":true,"dense":true},{"id":349,"key":"vision2web/current","label":"Vision2Web","capability":"vision","harness":"published","source_url":"https://benchlm.ai/benchmarks/vision2web","featured":false,"n_measured":4,"n_estimated":0,"listed":false,"dense":false},{"id":283,"key":"voxpopuliwer/current","label":"VoxPopuli WER","capability":"vision","harness":"published","source_url":"https://benchlm.ai/benchmarks/voxpopuliwer","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":143,"key":"vstar/current","label":"V*","capability":"vision","harness":"published","source_url":"https://benchlm.ai/benchmarks/vstar","featured":false,"n_measured":11,"n_estimated":16,"listed":true,"dense":true},{"id":310,"key":"worldvqaforceanswer/current","label":"WorldVQA ForceAnswer","capability":"vision","harness":"published","source_url":"https://benchlm.ai/benchmarks/worldvqaforceanswer","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":234,"key":"zerobench/current","label":"ZeroBench","capability":"vision","harness":"published","source_url":"https://benchlm.ai/benchmarks/zerobench","featured":false,"n_measured":8,"n_estimated":20,"listed":true,"dense":true},{"id":221,"key":"zerobenchpython/current","label":"ZeroBench w/ Python","capability":"vision","harness":"published","source_url":"https://benchlm.ai/benchmarks/zerobenchpython","featured":false,"n_measured":3,"n_estimated":0,"listed":false,"dense":false},{"id":67,"key":"aa-mlcr/current","label":"MLCR-AA","capability":"long-context","harness":"published","source_url":"https://benchlm.ai/benchmarks/aamlcr","featured":false,"n_measured":23,"n_estimated":0,"listed":false,"dense":false},{"id":223,"key":"corpusqa1m/current","label":"CorpusQA 1M","capability":"long-context","harness":"published","source_url":"https://benchlm.ai/benchmarks/corpusqa1m","featured":false,"n_measured":2,"n_estimated":0,"listed":false,"dense":false},{"id":249,"key":"graphwalksbfs128k/current","label":"Graphwalks BFS 128K","capability":"long-context","harness":"published","source_url":"https://benchlm.ai/benchmarks/graphwalksbfs128k","featured":false,"n_measured":2,"n_estimated":0,"listed":false,"dense":false},{"id":16,"key":"lcr/current","label":"AA-LCR","capability":"long-context","harness":"artificial-analysis","source_url":"https://benchlm.ai/benchmarks/lcr","featured":false,"n_measured":186,"n_estimated":8,"listed":true,"dense":true},{"id":336,"key":"long-context-translation/current","label":"Long-context translation","capability":"long-context","harness":"published","source_url":"https://benchlm.ai/benchmarks/long-context-translation","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":24,"key":"longbench-v2/current","label":"LongBench v2","capability":"long-context","harness":"published","source_url":"https://benchlm.ai/benchmarks/longbench-v2","featured":false,"n_measured":15,"n_estimated":176,"listed":true,"dense":true},{"id":334,"key":"mmlongbenchdoc/current","label":"MMLongBench-Doc","capability":"long-context","harness":"published","source_url":"https://benchlm.ai/benchmarks/mmlongbenchdoc","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":159,"key":"mrcr-v2-128k-256k/current","label":"MRCR v2 128K-256K","capability":"long-context","harness":"published","source_url":"https://benchlm.ai/benchmarks/mrcr-v2-128k-256k","featured":false,"n_measured":2,"n_estimated":0,"listed":false,"dense":false},{"id":272,"key":"mrcr-v2-256k-512k/current","label":"MRCR v2 256K-512K","capability":"long-context","harness":"published","source_url":"https://benchlm.ai/benchmarks/mrcr-v2-256k-512k","featured":false,"n_measured":2,"n_estimated":0,"listed":false,"dense":false},{"id":273,"key":"mrcr-v2-512k-1m/current","label":"MRCR v2 512K-1M","capability":"long-context","harness":"published","source_url":"https://benchlm.ai/benchmarks/mrcr-v2-512k-1m","featured":false,"n_measured":2,"n_estimated":0,"listed":false,"dense":false},{"id":246,"key":"mrcr-v2-64k-128k/current","label":"MRCR v2 64K-128K","capability":"long-context","harness":"published","source_url":"https://benchlm.ai/benchmarks/mrcr-v2-64k-128k","featured":false,"n_measured":2,"n_estimated":0,"listed":false,"dense":false},{"id":228,"key":"mrcr1m/current","label":"MRCR 1M","capability":"long-context","harness":"published","source_url":"https://benchlm.ai/benchmarks/mrcr1m","featured":false,"n_measured":4,"n_estimated":0,"listed":false,"dense":false},{"id":244,"key":"mrcrv2/current","label":"MRCRv2","capability":"long-context","harness":"published","source_url":"https://benchlm.ai/benchmarks/mrcrv2","featured":false,"n_measured":9,"n_estimated":0,"listed":false,"dense":false},{"id":304,"key":"acecyberrangesolved/current","label":"ACE solved","capability":"general","harness":"published","source_url":"https://benchlm.ai/benchmarks/acecyberrangesolved","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":265,"key":"advancedcybercompletionrateblue/current","label":"ACCR Daybreak Blue","capability":"general","harness":"published","source_url":"https://benchlm.ai/benchmarks/advancedcybercompletionrateblue","featured":false,"n_measured":2,"n_estimated":0,"listed":false,"dense":false},{"id":264,"key":"advancedcybercompletionratered/current","label":"ACCR Daybreak Red","capability":"general","harness":"published","source_url":"https://benchlm.ai/benchmarks/advancedcybercompletionratered","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":266,"key":"advancedcybercompletionratestandard/current","label":"ACCR standard","capability":"general","harness":"published","source_url":"https://benchlm.ai/benchmarks/advancedcybercompletionratestandard","featured":false,"n_measured":2,"n_estimated":0,"listed":false,"dense":false},{"id":110,"key":"anthropicossfuzzanycrash/current","label":"Anthropic OSS-Fuzz crash","capability":"general","harness":"published","source_url":"https://benchlm.ai/benchmarks/anthropicossfuzzanycrash","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":111,"key":"anthropicossfuzzwriteprimitive/current","label":"Anthropic OSS-Fuzz write primitive","capability":"general","harness":"published","source_url":"https://benchlm.ai/benchmarks/anthropicossfuzzwriteprimitive","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":261,"key":"atomicevasion/current","label":"Atomic evasion","capability":"general","harness":"published","source_url":"https://benchlm.ai/benchmarks/atomicevasion","featured":false,"n_measured":3,"n_estimated":0,"listed":false,"dense":false},{"id":262,"key":"atomicnetworkattacksimulation/current","label":"Atomic network attacks","capability":"general","harness":"published","source_url":"https://benchlm.ai/benchmarks/atomicnetworkattacksimulation","featured":false,"n_measured":3,"n_estimated":0,"listed":false,"dense":false},{"id":263,"key":"atomicvulnerabilityresearch/current","label":"Atomic vulnerability research","capability":"general","harness":"published","source_url":"https://benchlm.ai/benchmarks/atomicvulnerabilityresearch","featured":false,"n_measured":3,"n_estimated":0,"listed":false,"dense":false},{"id":284,"key":"bbh-perplexity-panel/current","label":"BBH (Perplexity panel)","capability":"general","harness":"published","source_url":"https://benchlm.ai/benchmarks/bbh-perplexity-panel","featured":false,"n_measured":3,"n_estimated":0,"listed":false,"dense":false},{"id":285,"key":"belebele-perplexity-panel/current","label":"Belebele (Perplexity panel)","capability":"general","harness":"published","source_url":"https://benchlm.ai/benchmarks/belebele-perplexity-panel","featured":false,"n_measured":3,"n_estimated":0,"listed":false,"dense":false},{"id":286,"key":"circa-perplexity-panel/current","label":"Circa (Perplexity panel)","capability":"general","harness":"published","source_url":"https://benchlm.ai/benchmarks/circa-perplexity-panel","featured":false,"n_measured":3,"n_estimated":0,"listed":false,"dense":false},{"id":238,"key":"cladder-decision-index-fastino/current","label":"CLadder (Fastino report)","capability":"general","harness":"published","source_url":"https://benchlm.ai/benchmarks/cladder-decision-index-fastino","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":287,"key":"contractnli-perplexity-panel/current","label":"ContractNLI (Perplexity panel)","capability":"general","harness":"published","source_url":"https://benchlm.ai/benchmarks/contractnli-perplexity-panel","featured":false,"n_measured":3,"n_estimated":0,"listed":false,"dense":false},{"id":239,"key":"cruxeval-decision-index-fastino/current","label":"CRUXEval (Fastino report)","capability":"general","harness":"published","source_url":"https://benchlm.ai/benchmarks/cruxeval-decision-index-fastino","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":267,"key":"cyscenariobenchaveragesuccess/current","label":"CyScenarioBench success","capability":"general","harness":"published","source_url":"https://benchlm.ai/benchmarks/cyscenariobenchaveragesuccess","featured":false,"n_measured":2,"n_estimated":0,"listed":false,"dense":false},{"id":268,"key":"cyscenariobenchscenariossolved/current","label":"CyScenarioBench solved","capability":"general","harness":"published","source_url":"https://benchlm.ai/benchmarks/cyscenariobenchscenariossolved","featured":false,"n_measured":2,"n_estimated":0,"listed":false,"dense":false},{"id":214,"key":"decision-index-0-2-1-fastino/current","label":"Decision Index 0.2.1 (Fastino report)","capability":"general","harness":"published","source_url":"https://benchlm.ai/benchmarks/decision-index-0-2-1-fastino","featured":false,"n_measured":6,"n_estimated":0,"listed":false,"dense":false},{"id":209,"key":"decision-index-0-2-1-liquid/current","label":"Decision Index 0.2.1 (Liquid report)","capability":"general","harness":"published","source_url":"https://benchlm.ai/benchmarks/decision-index-0-2-1-liquid","featured":false,"n_measured":2,"n_estimated":0,"listed":false,"dense":false},{"id":237,"key":"decision-index-arts-fastino/current","label":"Arts and Human Taste (Fastino report)","capability":"general","harness":"published","source_url":"https://benchlm.ai/benchmarks/decision-index-arts-fastino","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":208,"key":"decision-index-arts-liquid/current","label":"Arts (Liquid report)","capability":"general","harness":"published","source_url":"https://benchlm.ai/benchmarks/decision-index-arts-liquid","featured":false,"n_measured":2,"n_estimated":0,"listed":false,"dense":false},{"id":240,"key":"decision-index-knowledge-fastino/current","label":"Knowledge and Reasoning (Fastino report)","capability":"general","harness":"published","source_url":"https://benchlm.ai/benchmarks/decision-index-knowledge-fastino","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":210,"key":"decision-index-knowledge-liquid/current","label":"Knowledge (Liquid report)","capability":"general","harness":"published","source_url":"https://benchlm.ai/benchmarks/decision-index-knowledge-liquid","featured":false,"n_measured":2,"n_estimated":0,"listed":false,"dense":false},{"id":241,"key":"decision-index-language-fastino/current","label":"Language Understanding (Fastino report)","capability":"general","harness":"published","source_url":"https://benchlm.ai/benchmarks/decision-index-language-fastino","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":211,"key":"decision-index-language-liquid/current","label":"Language (Liquid report)","capability":"general","harness":"published","source_url":"https://benchlm.ai/benchmarks/decision-index-language-liquid","featured":false,"n_measured":2,"n_estimated":0,"listed":false,"dense":false},{"id":242,"key":"decision-index-retrieval-fastino/current","label":"Retrieval and Classification (Fastino report)","capability":"general","harness":"published","source_url":"https://benchlm.ai/benchmarks/decision-index-retrieval-fastino","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":212,"key":"decision-index-retrieval-liquid/current","label":"Retrieval (Liquid report)","capability":"general","harness":"published","source_url":"https://benchlm.ai/benchmarks/decision-index-retrieval-liquid","featured":false,"n_measured":2,"n_estimated":0,"listed":false,"dense":false},{"id":243,"key":"decision-index-tools-fastino/current","label":"Tools and Automation (Fastino report)","capability":"general","harness":"published","source_url":"https://benchlm.ai/benchmarks/decision-index-tools-fastino","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":213,"key":"decision-index-tools-liquid/current","label":"Tools (Liquid report)","capability":"general","harness":"published","source_url":"https://benchlm.ai/benchmarks/decision-index-tools-liquid","featured":false,"n_measured":2,"n_estimated":0,"listed":false,"dense":false},{"id":48,"key":"decisionindex/current","label":"Decision Index 0.3 Full Score","capability":"general","harness":"published","source_url":"https://benchlm.ai/benchmarks/decisionindex","featured":false,"n_measured":112,"n_estimated":0,"listed":false,"dense":false},{"id":114,"key":"exploitbench/current","label":"ExploitBench","capability":"general","harness":"published","source_url":"https://benchlm.ai/benchmarks/exploitbench","featured":false,"n_measured":11,"n_estimated":4,"listed":true,"dense":true},{"id":288,"key":"financialphrasebank-perplexity-panel/current","label":"FinancialPhraseBank (Perplexity panel)","capability":"general","harness":"published","source_url":"https://benchlm.ai/benchmarks/financialphrasebank-perplexity-panel","featured":false,"n_measured":3,"n_estimated":0,"listed":false,"dense":false},{"id":115,"key":"firefox147workingexploit/current","label":"Firefox 147 exploits","capability":"general","harness":"published","source_url":"https://benchlm.ai/benchmarks/firefox147workingexploit","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":269,"key":"frontiercyber/current","label":"FrontierCyber","capability":"general","harness":"published","source_url":"https://benchlm.ai/benchmarks/frontiercyber","featured":false,"n_measured":2,"n_estimated":0,"listed":false,"dense":false},{"id":289,"key":"jevbench-public-hard-perplexity-panel/current","label":"JevBench public hard (Perplexity panel)","capability":"general","harness":"published","source_url":"https://benchlm.ai/benchmarks/jevbench-public-hard-perplexity-panel","featured":false,"n_measured":3,"n_estimated":0,"listed":false,"dense":false},{"id":49,"key":"jevdecisionindex03arts/current","label":"Decision Index 0.3 Arts","capability":"general","harness":"published","source_url":"https://benchlm.ai/benchmarks/jevdecisionindex03arts","featured":false,"n_measured":112,"n_estimated":0,"listed":false,"dense":false},{"id":50,"key":"jevdecisionindex03knowledge/current","label":"Decision Index 0.3 Knowledge","capability":"general","harness":"published","source_url":"https://benchlm.ai/benchmarks/jevdecisionindex03knowledge","featured":false,"n_measured":112,"n_estimated":0,"listed":false,"dense":false},{"id":51,"key":"jevdecisionindex03language/current","label":"Decision Index 0.3 Language","capability":"general","harness":"published","source_url":"https://benchlm.ai/benchmarks/jevdecisionindex03language","featured":false,"n_measured":112,"n_estimated":0,"listed":false,"dense":false},{"id":52,"key":"jevdecisionindex03retrieval/current","label":"Decision Index 0.3 Retrieval","capability":"general","harness":"published","source_url":"https://benchlm.ai/benchmarks/jevdecisionindex03retrieval","featured":false,"n_measured":112,"n_estimated":0,"listed":false,"dense":false},{"id":53,"key":"jevdecisionindex03tools/current","label":"Decision Index 0.3 Tools","capability":"general","harness":"published","source_url":"https://benchlm.ai/benchmarks/jevdecisionindex03tools","featured":false,"n_measured":112,"n_estimated":0,"listed":false,"dense":false},{"id":290,"key":"judgebench-perplexity-panel/current","label":"JudgeBench (Perplexity panel)","capability":"general","harness":"published","source_url":"https://benchlm.ai/benchmarks/judgebench-perplexity-panel","featured":false,"n_measured":3,"n_estimated":0,"listed":false,"dense":false},{"id":116,"key":"lastonescyberrangecompletion/current","label":"The Last Ones completion","capability":"general","harness":"published","source_url":"https://benchlm.ai/benchmarks/lastonescyberrangecompletion","featured":false,"n_measured":3,"n_estimated":0,"listed":false,"dense":false},{"id":306,"key":"lastonescyberrangesteps/current","label":"The Last Ones steps","capability":"general","harness":"published","source_url":"https://benchlm.ai/benchmarks/lastonescyberrangesteps","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":291,"key":"perplexity-decision-panel/current","label":"Perplexity Decision Panel","capability":"general","harness":"published","source_url":"https://benchlm.ai/benchmarks/perplexity-decision-panel","featured":false,"n_measured":3,"n_estimated":0,"listed":false,"dense":false},{"id":292,"key":"ragtruth-perplexity-panel/current","label":"RAGTruth (Perplexity panel)","capability":"general","harness":"published","source_url":"https://benchlm.ai/benchmarks/ragtruth-perplexity-panel","featured":false,"n_measured":3,"n_estimated":0,"listed":false,"dense":false},{"id":119,"key":"sconepostcutoffsuccess/current","label":"SCONE post-cutoff success","capability":"general","harness":"published","source_url":"https://benchlm.ai/benchmarks/sconepostcutoffsuccess","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":219,"key":"secbenchpro/current","label":"SEC-Bench Pro","capability":"general","harness":"published","source_url":"https://benchlm.ai/benchmarks/secbenchpro","featured":false,"n_measured":9,"n_estimated":6,"listed":true,"dense":true},{"id":274,"key":"srebench/current","label":"SRE-Bench","capability":"general","harness":"published","source_url":"https://benchlm.ai/benchmarks/srebench","featured":false,"n_measured":1,"n_estimated":0,"listed":false,"dense":false},{"id":293,"key":"tabfact-perplexity-panel/current","label":"TabFact (Perplexity panel)","capability":"general","harness":"published","source_url":"https://benchlm.ai/benchmarks/tabfact-perplexity-panel","featured":false,"n_measured":3,"n_estimated":0,"listed":false,"dense":false},{"id":294,"key":"truthfulqa-binary-perplexity-panel/current","label":"TruthfulQA binary (Perplexity panel)","capability":"general","harness":"published","source_url":"https://benchlm.ai/benchmarks/truthfulqa-binary-perplexity-panel","featured":false,"n_measured":3,"n_estimated":0,"listed":false,"dense":false},{"id":339,"key":"vidore-v3-image-perplexity/current","label":"ViDoRe V3 Image (Perplexity report)","capability":"general","harness":"published","source_url":"https://benchlm.ai/benchmarks/vidore-v3-image-perplexity","featured":false,"n_measured":2,"n_estimated":0,"listed":false,"dense":false},{"id":340,"key":"vidore-v3-markdown-perplexity/current","label":"ViDoRe V3 Markdown (Perplexity report)","capability":"general","harness":"published","source_url":"https://benchlm.ai/benchmarks/vidore-v3-markdown-perplexity","featured":false,"n_measured":2,"n_estimated":0,"listed":false,"dense":false},{"id":295,"key":"winogrande-perplexity-panel/current","label":"WinoGrande (Perplexity panel)","capability":"general","harness":"published","source_url":"https://benchlm.ai/benchmarks/winogrande-perplexity-panel","featured":false,"n_measured":3,"n_estimated":0,"listed":false,"dense":false}],"models":[{"id":1,"slug":"a-x-k2","name":"A.X K2","provider":"other","dense":true,"listed":true,"n_measured":15,"n_estimated":75},{"id":63,"slug":"decision-index-0-3-ajev-lora5-gemma-4-12b","name":"AJev lora5 (Gemma 4 12B)","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":2,"slug":"agents-a1","name":"Agents-A1","provider":"other","dense":true,"listed":true,"n_measured":6,"n_estimated":44},{"id":3,"slug":"agents-a1-4b","name":"Agents-A1-4B","provider":"other","dense":true,"listed":true,"n_measured":8,"n_estimated":38},{"id":278,"slug":"gte-reranker-modernbert-base","name":"Alibaba GTE Reranker ModernBERT-base","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":4,"slug":"apodex-1-1","name":"Apodex 1.1","provider":"other","dense":true,"listed":true,"n_measured":18,"n_estimated":83},{"id":5,"slug":"apodex-1-1-mini","name":"Apodex 1.1 Mini","provider":"other","dense":true,"listed":true,"n_measured":12,"n_estimated":81},{"id":6,"slug":"atria-dawn-preview","name":"Atria Dawn Preview","provider":"other","dense":true,"listed":true,"n_measured":11,"n_estimated":46},{"id":7,"slug":"autojev-27b","name":"AutoJev-27B","provider":"other","dense":false,"listed":true,"n_measured":0,"n_estimated":1},{"id":159,"slug":"decision-index-autojev-27b","name":"AutoJev-27B (Decision Index reference)","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":8,"slug":"autojev-27b-rtxpro6000-a2","name":"AutoJev-27B (RTX PRO 6000)","provider":"other","dense":false,"listed":true,"n_measured":0,"n_estimated":1},{"id":9,"slug":"autoloops-gemma-4-31b-it","name":"Autoloops Gemma 4 31B IT","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":11,"slug":"bge-reranker-v2-m3","name":"BAAI bge-reranker-v2-m3","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":13,"slug":"btl-3","name":"BTL-3","provider":"other","dense":false,"listed":true,"n_measured":2,"n_estimated":9},{"id":14,"slug":"btl-4","name":"BTL-4","provider":"other","dense":true,"listed":true,"n_measured":3,"n_estimated":14},{"id":477,"slug":"reflection-beam","name":"Beam","provider":"other","dense":true,"listed":true,"n_measured":14,"n_estimated":81},{"id":64,"slug":"decision-index-0-3-bekko-system-one-v0-17m","name":"Bekko System One v0 17M","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":65,"slug":"decision-index-0-3-bekko-system-one-v0-400m","name":"Bekko System One v0 400M","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":66,"slug":"decision-index-0-3-bekko-system-one-v0-68m","name":"Bekko System One v0 68M","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":402,"slug":"nimble-9b","name":"Bespoke Nimble 9B","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":67,"slug":"decision-index-0-3-bespoke-nimble-9b-v2","name":"Bespoke Nimble 9B v2","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":68,"slug":"decision-index-0-3-bespoke-nimble-9b-v3","name":"Bespoke Nimble 9B v3","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":10,"slug":"bev-bonsai-27b","name":"Bev / Bonsai 27B","provider":"other","dense":false,"listed":true,"n_measured":0,"n_estimated":1},{"id":69,"slug":"decision-index-0-3-blink-v0-3-26b-a4b-nvfp4","name":"Blink v0.3 26B-A4B (NVFP4)","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":12,"slug":"bosun-v31-0-6b","name":"Bosun v3.1 0.6B","provider":"other","dense":false,"listed":true,"n_measured":0,"n_estimated":1},{"id":70,"slug":"decision-index-0-3-bosun-v3-1-1-7b","name":"Bosun v3.1 1.7B","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":46,"slug":"clm-8b","name":"CLM-8B","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":71,"slug":"decision-index-0-3-clm-v0-1-8b","name":"CLM-v0.1-8B","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":15,"slug":"celeris-1","name":"Celeris-1","provider":"other","dense":true,"listed":true,"n_measured":12,"n_estimated":73},{"id":16,"slug":"certo","name":"Certo v1","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":19,"slug":"claude-3-haiku","name":"Claude 3 Haiku","provider":"anthropic","dense":true,"listed":true,"n_measured":9,"n_estimated":46},{"id":20,"slug":"claude-3-opus","name":"Claude 3 Opus","provider":"anthropic","dense":true,"listed":true,"n_measured":4,"n_estimated":46},{"id":18,"slug":"claude-3-5-sonnet","name":"Claude 3.5 Sonnet","provider":"anthropic","dense":true,"listed":true,"n_measured":4,"n_estimated":36},{"id":23,"slug":"claude-4-sonnet","name":"Claude 4 Sonnet","provider":"anthropic","dense":true,"listed":true,"n_measured":12,"n_estimated":74},{"id":21,"slug":"claude-4-1-opus","name":"Claude 4.1 Opus","provider":"anthropic","dense":true,"listed":true,"n_measured":3,"n_estimated":52},{"id":22,"slug":"claude-4-1-opus-thinking","name":"Claude 4.1 Opus Thinking","provider":"anthropic","dense":true,"listed":true,"n_measured":8,"n_estimated":47},{"id":24,"slug":"claude-fable","name":"Claude Fable 5","provider":"anthropic","dense":true,"listed":true,"n_measured":34,"n_estimated":67},{"id":25,"slug":"claude-fable-5-1","name":"Claude Fable 5.1","provider":"anthropic","dense":true,"listed":true,"n_measured":41,"n_estimated":61},{"id":26,"slug":"claude-haiku-4-5","name":"Claude Haiku 4.5","provider":"anthropic","dense":true,"listed":true,"n_measured":10,"n_estimated":59},{"id":27,"slug":"claude-haiku-4-5-thinking","name":"Claude Haiku 4.5 Thinking","provider":"anthropic","dense":false,"listed":true,"n_measured":1,"n_estimated":8},{"id":28,"slug":"claude-haiku-5-5","name":"Claude Haiku 5.5","provider":"anthropic","dense":true,"listed":true,"n_measured":13,"n_estimated":65},{"id":29,"slug":"claude-mythos-5","name":"Claude Mythos 5","provider":"anthropic","dense":true,"listed":true,"n_measured":13,"n_estimated":82},{"id":30,"slug":"claude-mythos-5-1","name":"Claude Mythos 5.1","provider":"anthropic","dense":false,"listed":true,"n_measured":1,"n_estimated":2},{"id":31,"slug":"claude-mythos-preview","name":"Claude Mythos Preview","provider":"anthropic","dense":true,"listed":true,"n_measured":3,"n_estimated":20},{"id":32,"slug":"claude-opus-4-5","name":"Claude Opus 4.5","provider":"anthropic","dense":true,"listed":true,"n_measured":52,"n_estimated":76},{"id":33,"slug":"claude-opus-4-5-thinking","name":"Claude Opus 4.5 Thinking","provider":"anthropic","dense":true,"listed":true,"n_measured":12,"n_estimated":52},{"id":34,"slug":"claude-opus-4-6","name":"Claude Opus 4.6","provider":"anthropic","dense":true,"listed":true,"n_measured":41,"n_estimated":76},{"id":35,"slug":"claude-opus-4-6-thinking","name":"Claude Opus 4.6 (Adaptive)","provider":"anthropic","dense":true,"listed":true,"n_measured":13,"n_estimated":58},{"id":36,"slug":"claude-opus-4-7","name":"Claude Opus 4.7","provider":"anthropic","dense":true,"listed":true,"n_measured":24,"n_estimated":86},{"id":37,"slug":"claude-opus-4-7-adaptive","name":"Claude Opus 4.7 (Adaptive)","provider":"anthropic","dense":true,"listed":true,"n_measured":32,"n_estimated":87},{"id":38,"slug":"claude-opus-4-8","name":"Claude Opus 4.8","provider":"anthropic","dense":true,"listed":true,"n_measured":47,"n_estimated":81},{"id":39,"slug":"claude-opus-5","name":"Claude Opus 5","provider":"anthropic","dense":true,"listed":true,"n_measured":54,"n_estimated":62},{"id":40,"slug":"claude-opus-5-5","name":"Claude Opus 5.5","provider":"anthropic","dense":true,"listed":true,"n_measured":40,"n_estimated":69},{"id":41,"slug":"claude-sonnet-4-5","name":"Claude Sonnet 4.5","provider":"anthropic","dense":true,"listed":true,"n_measured":11,"n_estimated":59},{"id":42,"slug":"claude-sonnet-4-5-thinking","name":"Claude Sonnet 4.5 Thinking","provider":"anthropic","dense":true,"listed":true,"n_measured":3,"n_estimated":24},{"id":43,"slug":"claude-sonnet-4-6","name":"Claude Sonnet 4.6","provider":"anthropic","dense":true,"listed":true,"n_measured":37,"n_estimated":82},{"id":44,"slug":"claude-sonnet-5","name":"Claude Sonnet 5","provider":"anthropic","dense":true,"listed":true,"n_measured":34,"n_estimated":83},{"id":45,"slug":"claude-sonnet-5-5","name":"Claude Sonnet 5.5","provider":"anthropic","dense":true,"listed":true,"n_measured":33,"n_estimated":65},{"id":47,"slug":"cloudflare-clef","name":"Clef","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":48,"slug":"cloudflare-clef-flash","name":"Clef-Flash","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":49,"slug":"command-a-plus","name":"Command A+","provider":"other","dense":true,"listed":true,"n_measured":17,"n_estimated":95},{"id":50,"slug":"composer-2","name":"Composer 2","provider":"other","dense":true,"listed":true,"n_measured":4,"n_estimated":9},{"id":52,"slug":"composer-2-fast","name":"Composer 2 Fast","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":5},{"id":51,"slug":"composer-2-5","name":"Composer 2.5","provider":"other","dense":true,"listed":true,"n_measured":7,"n_estimated":22},{"id":53,"slug":"cygnet","name":"Cygnet","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":78,"slug":"decision-index-0-3-dm-jepa-1-1","name":"DM-JEPA 1.1","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":72,"slug":"decision-index-0-3-decider-4b","name":"Decider 4B","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":160,"slug":"decision-index-decider-chat-gemma4-31b","name":"Decider chat · Gemma-4-31B","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":73,"slug":"decision-index-0-3-decider-chat-qwen3-6-27b","name":"Decider chat · Qwen3.6-27B","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":59,"slug":"decision-2b","name":"Decision 2B v59","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":60,"slug":"decision-4b-v11","name":"Decision 4B v1.1","provider":"other","dense":false,"listed":true,"n_measured":0,"n_estimated":1},{"id":61,"slug":"decision-4b-v12","name":"Decision 4B v1.2","provider":"other","dense":false,"listed":true,"n_measured":0,"n_estimated":1},{"id":62,"slug":"decision-fast","name":"Decision Fast v53a","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":164,"slug":"deem-0-8-v1","name":"Deem 0.8B v1","provider":"other","dense":false,"listed":true,"n_measured":0,"n_estimated":1},{"id":166,"slug":"deepseek-r1-distill-qwen-32b","name":"DeepSeek R1 Distill Qwen 32B","provider":"deepseek","dense":true,"listed":true,"n_measured":5,"n_estimated":29},{"id":167,"slug":"deepseek-v3","name":"DeepSeek V3","provider":"deepseek","dense":true,"listed":true,"n_measured":18,"n_estimated":75},{"id":168,"slug":"deepseek-v3-0324","name":"DeepSeek V3 0324","provider":"deepseek","dense":true,"listed":true,"n_measured":11,"n_estimated":71},{"id":169,"slug":"deepseek-v3-1","name":"DeepSeek V3.1","provider":"deepseek","dense":true,"listed":true,"n_measured":8,"n_estimated":34},{"id":170,"slug":"deepseek-v3-1-reasoning","name":"DeepSeek V3.1 (Reasoning)","provider":"deepseek","dense":true,"listed":true,"n_measured":8,"n_estimated":34},{"id":171,"slug":"deepseek-v3-2","name":"DeepSeek V3.2","provider":"deepseek","dense":true,"listed":true,"n_measured":15,"n_estimated":57},{"id":172,"slug":"deepseek-v3-2-thinking","name":"DeepSeek V3.2 (Thinking)","provider":"deepseek","dense":false,"listed":true,"n_measured":1,"n_estimated":8},{"id":174,"slug":"deepseek-v4-flash-0731","name":"DeepSeek V4 Flash 0731","provider":"deepseek","dense":true,"listed":true,"n_measured":40,"n_estimated":70},{"id":175,"slug":"deepseek-v4-pro-0813","name":"DeepSeek V4 Pro 0813","provider":"deepseek","dense":true,"listed":true,"n_measured":44,"n_estimated":69},{"id":173,"slug":"deepseek-v4-1-flash","name":"DeepSeek V4.1 Flash","provider":"deepseek","dense":true,"listed":true,"n_measured":30,"n_estimated":79},{"id":165,"slug":"deepseek-r1","name":"DeepSeek-R1","provider":"deepseek","dense":true,"listed":true,"n_measured":8,"n_estimated":34},{"id":77,"slug":"decision-index-0-3-dinah-0","name":"Dinah-0","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":79,"slug":"decision-index-0-3-exaone-4-0-1-2b-jev-v0-3","name":"EXAONE-4.0-1.2B-JEV v0.3","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":179,"slug":"eikos-27b","name":"Eikos-27B","provider":"other","dense":false,"listed":true,"n_measured":0,"n_estimated":1},{"id":180,"slug":"ember-1","name":"Ember-1","provider":"other","dense":true,"listed":true,"n_measured":3,"n_estimated":18},{"id":181,"slug":"exaone-4-0-1-2b","name":"Exaone 4.0 1.2B","provider":"other","dense":true,"listed":true,"n_measured":8,"n_estimated":34},{"id":182,"slug":"exaone-4-0-32b","name":"Exaone 4.0 32B","provider":"other","dense":true,"listed":true,"n_measured":9,"n_estimated":33},{"id":183,"slug":"fara-1-5-27b","name":"Fara1.5-27B","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":184,"slug":"fara-1-5-4b","name":"Fara1.5-4B","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":81,"slug":"decision-index-0-3-fastino-glide-no-thinking-28b","name":"Fastino GLiDE no-thinking (28B)","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":481,"slug":"sakana-fugu-cyber","name":"Fugu Cyber","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":15},{"id":82,"slug":"decision-index-0-3-gev-26b-decide","name":"GEV-26B-Decide","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":213,"slug":"glm-4-5","name":"GLM-4.5","provider":"zhipu","dense":true,"listed":true,"n_measured":3,"n_estimated":23},{"id":214,"slug":"glm-4-5-air","name":"GLM-4.5-Air","provider":"zhipu","dense":true,"listed":true,"n_measured":8,"n_estimated":34},{"id":215,"slug":"glm-4-6","name":"GLM-4.6","provider":"zhipu","dense":true,"listed":true,"n_measured":14,"n_estimated":48},{"id":216,"slug":"glm-4-7","name":"GLM-4.7","provider":"zhipu","dense":true,"listed":true,"n_measured":27,"n_estimated":76},{"id":217,"slug":"glm-5","name":"GLM-5","provider":"zhipu","dense":true,"listed":true,"n_measured":42,"n_estimated":66},{"id":222,"slug":"glm-5-reasoning","name":"GLM-5 (Reasoning)","provider":"zhipu","dense":false,"listed":true,"n_measured":1,"n_estimated":8},{"id":223,"slug":"glm-5-turbo","name":"GLM-5-Turbo","provider":"zhipu","dense":true,"listed":true,"n_measured":9,"n_estimated":40},{"id":218,"slug":"glm-5-1","name":"GLM-5.1","provider":"zhipu","dense":true,"listed":true,"n_measured":38,"n_estimated":77},{"id":219,"slug":"glm-5-2","name":"GLM-5.2","provider":"zhipu","dense":true,"listed":true,"n_measured":38,"n_estimated":76},{"id":220,"slug":"glm-5-3","name":"GLM-5.3","provider":"zhipu","dense":true,"listed":true,"n_measured":36,"n_estimated":63},{"id":221,"slug":"glm-5-3-flash","name":"GLM-5.3-Flash","provider":"zhipu","dense":true,"listed":true,"n_measured":31,"n_estimated":69},{"id":224,"slug":"glm-5v-turbo","name":"GLM-5V-Turbo","provider":"zhipu","dense":true,"listed":true,"n_measured":11,"n_estimated":58},{"id":185,"slug":"fastino-glide","name":"GLiDE","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":212,"slug":"gliner2-large","name":"GLiNER2 large","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":209,"slug":"gliner2-5-base","name":"GLiNER2.5 base","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":210,"slug":"gliner2-5-multi","name":"GLiNER2.5 multi","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":211,"slug":"gliner2-5-small","name":"GLiNER2.5 small","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":85,"slug":"decision-index-0-3-gliner2-5-decide","name":"GLiNER2.5-Decide","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":228,"slug":"gpt-4-turbo","name":"GPT-4 Turbo","provider":"openai","dense":true,"listed":true,"n_measured":3,"n_estimated":47},{"id":225,"slug":"gpt-4-1","name":"GPT-4.1","provider":"openai","dense":true,"listed":true,"n_measured":17,"n_estimated":72},{"id":226,"slug":"gpt-4-1-mini","name":"GPT-4.1 mini","provider":"openai","dense":true,"listed":true,"n_measured":16,"n_estimated":87},{"id":227,"slug":"gpt-4-1-nano","name":"GPT-4.1 nano","provider":"openai","dense":true,"listed":true,"n_measured":15,"n_estimated":87},{"id":229,"slug":"gpt-4o","name":"GPT-4o","provider":"openai","dense":true,"listed":true,"n_measured":11,"n_estimated":60},{"id":230,"slug":"gpt-4o-mini","name":"GPT-4o mini","provider":"openai","dense":true,"listed":true,"n_measured":7,"n_estimated":85},{"id":248,"slug":"gpt-5-high","name":"GPT-5 (high)","provider":"openai","dense":true,"listed":true,"n_measured":13,"n_estimated":87},{"id":249,"slug":"gpt-5-medium","name":"GPT-5 (medium)","provider":"openai","dense":true,"listed":true,"n_measured":9,"n_estimated":46},{"id":250,"slug":"gpt-5-mini","name":"GPT-5 mini","provider":"openai","dense":false,"listed":true,"n_measured":1,"n_estimated":8},{"id":231,"slug":"gpt-5-1","name":"GPT-5.1","provider":"openai","dense":true,"listed":true,"n_measured":16,"n_estimated":91},{"id":232,"slug":"gpt-5-1-codex","name":"GPT-5.1-Codex","provider":"openai","dense":true,"listed":true,"n_measured":13,"n_estimated":74},{"id":233,"slug":"gpt-5-1-codex-max","name":"GPT-5.1-Codex-Max","provider":"openai","dense":true,"listed":true,"n_measured":11,"n_estimated":58},{"id":234,"slug":"gpt-5-2","name":"GPT-5.2","provider":"openai","dense":true,"listed":true,"n_measured":24,"n_estimated":91},{"id":235,"slug":"gpt-5-2-codex","name":"GPT-5.2-Codex","provider":"openai","dense":true,"listed":true,"n_measured":14,"n_estimated":80},{"id":236,"slug":"gpt-5-3-codex","name":"GPT-5.3 Codex","provider":"openai","dense":true,"listed":true,"n_measured":19,"n_estimated":80},{"id":237,"slug":"gpt-5-4","name":"GPT-5.4","provider":"openai","dense":true,"listed":true,"n_measured":51,"n_estimated":73},{"id":240,"slug":"gpt-5-4-pro","name":"GPT-5.4 Pro","provider":"openai","dense":true,"listed":true,"n_measured":9,"n_estimated":50},{"id":238,"slug":"gpt-5-4-mini","name":"GPT-5.4 mini","provider":"openai","dense":true,"listed":true,"n_measured":34,"n_estimated":89},{"id":239,"slug":"gpt-5-4-nano","name":"GPT-5.4 nano","provider":"openai","dense":true,"listed":true,"n_measured":33,"n_estimated":90},{"id":241,"slug":"gpt-5-5","name":"GPT-5.5","provider":"openai","dense":true,"listed":true,"n_measured":54,"n_estimated":75},{"id":242,"slug":"gpt-5-5-pro","name":"GPT-5.5 Pro","provider":"openai","dense":true,"listed":true,"n_measured":9,"n_estimated":50},{"id":243,"slug":"gpt-5-6-cyber","name":"GPT-5.6 Cyber","provider":"openai","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":244,"slug":"gpt-5-6-luna","name":"GPT-5.6 Luna","provider":"openai","dense":true,"listed":true,"n_measured":43,"n_estimated":82},{"id":245,"slug":"gpt-5-6-luna-low","name":"GPT-5.6 Luna (low)","provider":"openai","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":246,"slug":"gpt-5-6-sol","name":"GPT-5.6 Sol","provider":"openai","dense":true,"listed":true,"n_measured":52,"n_estimated":76},{"id":247,"slug":"gpt-5-6-terra","name":"GPT-5.6 Terra","provider":"openai","dense":true,"listed":true,"n_measured":49,"n_estimated":79},{"id":252,"slug":"gpt-6-astra","name":"GPT-6 Astra","provider":"openai","dense":true,"listed":true,"n_measured":48,"n_estimated":75},{"id":253,"slug":"gpt-6-luna","name":"GPT-6 Luna","provider":"openai","dense":true,"listed":true,"n_measured":22,"n_estimated":77},{"id":254,"slug":"gpt-6-luna-low","name":"GPT-6 Luna (low)","provider":"openai","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":255,"slug":"gpt-6-luna-medium","name":"GPT-6 Luna (medium)","provider":"openai","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":256,"slug":"gpt-6-sol","name":"GPT-6 Sol","provider":"openai","dense":true,"listed":true,"n_measured":27,"n_estimated":74},{"id":251,"slug":"gpt-6-1-sol","name":"GPT-6.1 Sol","provider":"openai","dense":true,"listed":true,"n_measured":26,"n_estimated":75},{"id":257,"slug":"gpt-oss-120b","name":"GPT-OSS 120B","provider":"openai","dense":true,"listed":true,"n_measured":15,"n_estimated":76},{"id":258,"slug":"gpt-oss-20b","name":"GPT-OSS 20B","provider":"openai","dense":true,"listed":true,"n_measured":14,"n_estimated":75},{"id":186,"slug":"gemini-1-0-pro","name":"Gemini 1.0 Pro","provider":"google","dense":true,"listed":true,"n_measured":3,"n_estimated":26},{"id":187,"slug":"gemini-1-5-pro","name":"Gemini 1.5 Pro","provider":"google","dense":true,"listed":true,"n_measured":5,"n_estimated":58},{"id":188,"slug":"gemini-2-5-flash","name":"Gemini 2.5 Flash","provider":"google","dense":true,"listed":true,"n_measured":11,"n_estimated":50},{"id":189,"slug":"gemini-2-5-pro","name":"Gemini 2.5 Pro","provider":"google","dense":true,"listed":true,"n_measured":21,"n_estimated":89},{"id":199,"slug":"gemini-3-flash","name":"Gemini 3 Flash","provider":"google","dense":true,"listed":true,"n_measured":20,"n_estimated":87},{"id":200,"slug":"gemini-3-pro","name":"Gemini 3 Pro","provider":"google","dense":true,"listed":true,"n_measured":22,"n_estimated":74},{"id":201,"slug":"gemini-3-pro-deep-think","name":"Gemini 3 Pro Deep Think","provider":"google","dense":false,"listed":true,"n_measured":2,"n_estimated":21},{"id":190,"slug":"gemini-3-1-flash-lite","name":"Gemini 3.1 Flash-Lite","provider":"google","dense":true,"listed":true,"n_measured":9,"n_estimated":64},{"id":191,"slug":"gemini-3-1-pro","name":"Gemini 3.1 Pro","provider":"google","dense":true,"listed":true,"n_measured":42,"n_estimated":85},{"id":192,"slug":"gemini-3-5-flash","name":"Gemini 3.5 Flash","provider":"google","dense":true,"listed":true,"n_measured":41,"n_estimated":86},{"id":193,"slug":"gemini-3-5-flash-cyber","name":"Gemini 3.5 Flash Cyber","provider":"google","dense":false,"listed":true,"n_measured":1,"n_estimated":15},{"id":194,"slug":"gemini-3-5-flash-lite","name":"Gemini 3.5 Flash-Lite","provider":"google","dense":true,"listed":true,"n_measured":19,"n_estimated":87},{"id":195,"slug":"gemini-3-6-flash","name":"Gemini 3.6 Flash","provider":"google","dense":true,"listed":true,"n_measured":21,"n_estimated":85},{"id":196,"slug":"gemini-3-7-flash","name":"Gemini 3.7 Flash","provider":"google","dense":true,"listed":true,"n_measured":35,"n_estimated":76},{"id":197,"slug":"gemini-3-8-flash","name":"Gemini 3.8 Flash","provider":"google","dense":true,"listed":true,"n_measured":38,"n_estimated":73},{"id":198,"slug":"gemini-3-8-flash-cyber","name":"Gemini 3.8 Flash Cyber","provider":"google","dense":false,"listed":true,"n_measured":1,"n_estimated":15},{"id":202,"slug":"gemini-4-argon","name":"Gemini 4 Argon","provider":"google","dense":true,"listed":true,"n_measured":20,"n_estimated":67},{"id":203,"slug":"gemma-3-27b","name":"Gemma 3 27B","provider":"other","dense":true,"listed":true,"n_measured":13,"n_estimated":87},{"id":204,"slug":"gemma-4-12b","name":"Gemma 4 12B","provider":"other","dense":true,"listed":true,"n_measured":21,"n_estimated":92},{"id":205,"slug":"gemma-4-26b-a4b","name":"Gemma 4 26B A4B","provider":"other","dense":true,"listed":true,"n_measured":16,"n_estimated":85},{"id":206,"slug":"gemma-4-31b","name":"Gemma 4 31B","provider":"other","dense":true,"listed":true,"n_measured":21,"n_estimated":87},{"id":207,"slug":"gemma-4-e2b","name":"Gemma 4 E2B","provider":"other","dense":true,"listed":true,"n_measured":13,"n_estimated":84},{"id":208,"slug":"gemma-4-e4b","name":"Gemma 4 E4B","provider":"other","dense":true,"listed":true,"n_measured":14,"n_estimated":83},{"id":83,"slug":"decision-index-0-3-gevva-e2b","name":"Gevva E2B","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":84,"slug":"decision-index-0-3-gevva-e4b","name":"Gevva E4B","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":262,"slug":"granite-4-2-30b","name":"Granite 4.2 30B","provider":"other","dense":true,"listed":true,"n_measured":21,"n_estimated":80},{"id":263,"slug":"granite-4-2-3b","name":"Granite 4.2 3B","provider":"other","dense":true,"listed":true,"n_measured":17,"n_estimated":71},{"id":264,"slug":"granite-4-2-8b","name":"Granite 4.2 8B","provider":"other","dense":true,"listed":true,"n_measured":23,"n_estimated":81},{"id":259,"slug":"granite-4-0-350m","name":"Granite-4.0-350M","provider":"other","dense":true,"listed":true,"n_measured":8,"n_estimated":34},{"id":260,"slug":"granite-4-0-h-1b","name":"Granite-4.0-H-1B","provider":"other","dense":true,"listed":true,"n_measured":8,"n_estimated":34},{"id":261,"slug":"granite-4-0-h-350m","name":"Granite-4.0-H-350M","provider":"other","dense":true,"listed":true,"n_measured":8,"n_estimated":34},{"id":265,"slug":"grok-3-beta","name":"Grok 3 [Beta]","provider":"xai","dense":false,"listed":true,"n_measured":2,"n_estimated":4},{"id":266,"slug":"grok-4","name":"Grok 4","provider":"xai","dense":true,"listed":true,"n_measured":13,"n_estimated":66},{"id":275,"slug":"grok-4-fast-reasoning","name":"Grok 4 Fast (Reasoning)","provider":"xai","dense":true,"listed":true,"n_measured":10,"n_estimated":54},{"id":267,"slug":"grok-4-1","name":"Grok 4.1","provider":"xai","dense":false,"listed":true,"n_measured":1,"n_estimated":10},{"id":268,"slug":"grok-4-1-fast","name":"Grok 4.1 Fast","provider":"xai","dense":true,"listed":true,"n_measured":10,"n_estimated":57},{"id":269,"slug":"grok-4-1-fast-reasoning","name":"Grok 4.1 Fast (Reasoning)","provider":"xai","dense":true,"listed":true,"n_measured":10,"n_estimated":54},{"id":270,"slug":"grok-4-20-beta","name":"Grok 4.20","provider":"xai","dense":true,"listed":true,"n_measured":23,"n_estimated":66},{"id":271,"slug":"grok-4-3","name":"Grok 4.3","provider":"xai","dense":true,"listed":true,"n_measured":26,"n_estimated":88},{"id":272,"slug":"grok-4-5","name":"Grok 4.5","provider":"xai","dense":true,"listed":true,"n_measured":27,"n_estimated":81},{"id":273,"slug":"grok-4-6","name":"Grok 4.6","provider":"xai","dense":true,"listed":true,"n_measured":31,"n_estimated":63},{"id":274,"slug":"grok-4-7","name":"Grok 4.7","provider":"xai","dense":true,"listed":true,"n_measured":19,"n_estimated":72},{"id":276,"slug":"grok-build-0-1","name":"Grok Build 0.1","provider":"xai","dense":false,"listed":true,"n_measured":1,"n_estimated":14},{"id":277,"slug":"grok-code-fast-1","name":"Grok Code Fast 1","provider":"xai","dense":true,"listed":true,"n_measured":10,"n_estimated":47},{"id":279,"slug":"holo2-235b-a22b","name":"Holo2-235B-A22B","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":6},{"id":280,"slug":"holo2-30b-a3b","name":"Holo2-30B-A3B","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":6},{"id":281,"slug":"holo2-4b","name":"Holo2-4B","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":6},{"id":282,"slug":"holo2-8b","name":"Holo2-8B","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":6},{"id":286,"slug":"holo3-122b-a10b","name":"Holo3-122B-A10B","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":12},{"id":287,"slug":"holo3-35b-a3b","name":"Holo3-35B-A3B","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":12},{"id":283,"slug":"holo3-1-35b-a3b","name":"Holo3.1-35B-A3B","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":284,"slug":"holo3-1-4b","name":"Holo3.1-4B","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":285,"slug":"holo3-1-9b","name":"Holo3.1-9B","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":288,"slug":"hopper","name":"Hopper","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":87,"slug":"decision-index-0-3-hopper-g-1-2","name":"Hopper (G) 1.2","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":86,"slug":"decision-index-0-3-hopper-12b-trained","name":"Hopper 12B (trained)","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":289,"slug":"hy3","name":"Hy3","provider":"other","dense":true,"listed":true,"n_measured":10,"n_estimated":72},{"id":290,"slug":"hy3-preview","name":"Hy3 Preview","provider":"other","dense":true,"listed":true,"n_measured":18,"n_estimated":75},{"id":291,"slug":"hy4-preview","name":"Hy4 preview","provider":"other","dense":true,"listed":true,"n_measured":25,"n_estimated":69},{"id":292,"slug":"imajev-4b","name":"Imajev-4B","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":293,"slug":"imajev-4b-rtx5090-a2","name":"Imajev-4B (RTX 5090)","provider":"other","dense":false,"listed":true,"n_measured":0,"n_estimated":1},{"id":294,"slug":"inkling","name":"Inkling","provider":"other","dense":true,"listed":true,"n_measured":39,"n_estimated":89},{"id":295,"slug":"inkling-small","name":"Inkling-Small","provider":"other","dense":true,"listed":true,"n_measured":35,"n_estimated":92},{"id":297,"slug":"instinct-dual-4b","name":"Instinct Dual 4B","provider":"other","dense":false,"listed":true,"n_measured":0,"n_estimated":1},{"id":296,"slug":"instinct","name":"Instinct Qwen3.8-27B","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":88,"slug":"decision-index-0-3-intelif-qwen3-4b","name":"Intelif Qwen3-4B","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":298,"slug":"interfaze-beta","name":"Interfaze Beta","provider":"other","dense":true,"listed":true,"n_measured":5,"n_estimated":35},{"id":89,"slug":"decision-index-0-3-intern-decision-0-8b","name":"Intern-Decision-0.8B","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":90,"slug":"decision-index-0-3-intern-decision-2b","name":"Intern-Decision-2B","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":91,"slug":"decision-index-0-3-intern-decision-4b","name":"Intern-Decision-4B","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":92,"slug":"decision-index-0-3-jade","name":"JADE","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":303,"slug":"jev-qwen3-5-9b-base-nvfp4","name":"JEV Qwen3.5-9B Base NVFP4","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":96,"slug":"decision-index-0-3-jev-27b","name":"JEV-27B","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":97,"slug":"decision-index-0-3-jev-9b","name":"JEV-9B","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":109,"slug":"decision-index-0-3-jpt-0-8b","name":"JPT-0.8B","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":110,"slug":"decision-index-0-3-jpt-35b-a3b","name":"JPT-35B-A3B","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":111,"slug":"decision-index-0-3-jpt-4b","name":"JPT-4B","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":112,"slug":"decision-index-0-3-jpt-9b","name":"JPT-9B","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":93,"slug":"decision-index-0-3-jebadiah-27b","name":"Jebadiah 27B","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":94,"slug":"decision-index-0-3-jet-v6-2","name":"Jet v6.2","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":95,"slug":"decision-index-0-3-jev","name":"Jev","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":300,"slug":"jev-1-13-0","name":"Jev 1.13.0","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":302,"slug":"jev-omni","name":"Jev-Omni","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":98,"slug":"decision-index-0-3-jev-style-0-8b-v3","name":"Jev-Style 0.8B v3","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":99,"slug":"decision-index-0-3-jev-style-2b-v3","name":"Jev-Style 2B v3","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":304,"slug":"jevact","name":"JevAct","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":101,"slug":"decision-index-0-3-jevk5-9b-v0-3-3","name":"JevK5 9B v0.3.3","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":102,"slug":"decision-index-0-3-jevk5-lite","name":"JevK5 Lite","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":305,"slug":"jevk5-v0-2-0","name":"JevK5 v0.2.0","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":103,"slug":"decision-index-0-3-jevk5-v0-3","name":"JevK5 v0.3","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":306,"slug":"jevk5-v0-3-4b","name":"JevK5 v0.3 4B","provider":"other","dense":false,"listed":true,"n_measured":0,"n_estimated":1},{"id":307,"slug":"jevone","name":"JevOne","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":100,"slug":"decision-index-0-3-jevfire","name":"Jevfire","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":104,"slug":"decision-index-0-3-jevstral-8b","name":"Jevstral 8B","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":107,"slug":"decision-index-0-3-jobe","name":"Jobe","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":308,"slug":"jobe-qwen3-5-4b","name":"Jobe Qwen3.5-4B","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":108,"slug":"decision-index-0-3-joshuasp-diffusiongemma-open-jev","name":"JoshuaSP diffusiongemma (open-jev)","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":113,"slug":"decision-index-0-3-julia-1","name":"Julia 1","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":311,"slug":"k-exaone-2-0","name":"K-EXAONE 2.0","provider":"other","dense":true,"listed":true,"n_measured":20,"n_estimated":60},{"id":310,"slug":"k-exaone","name":"K-Exaone","provider":"other","dense":true,"listed":true,"n_measured":10,"n_estimated":74},{"id":312,"slug":"kanana-2-1-3b-instruct","name":"Kanana-2 1.3B Instruct","provider":"other","dense":false,"listed":true,"n_measured":3,"n_estimated":1},{"id":313,"slug":"kanana-2-3b-instruct","name":"Kanana-2 3B Instruct","provider":"other","dense":false,"listed":true,"n_measured":3,"n_estimated":1},{"id":114,"slug":"decision-index-0-3-kas-4b","name":"Kas 4B","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":115,"slug":"decision-index-0-3-kev-0-8b-r15","name":"Kev 0.8B r15","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":116,"slug":"decision-index-0-3-kev-27b","name":"Kev 27B","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":117,"slug":"decision-index-0-3-kev-4b-r10","name":"Kev 4B r10","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":118,"slug":"decision-index-0-3-kev-9b-v2","name":"Kev 9B v2","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":319,"slug":"kimi-k2","name":"Kimi K2","provider":"moonshot","dense":true,"listed":true,"n_measured":10,"n_estimated":38},{"id":320,"slug":"kimi-k2-5","name":"Kimi K2.5","provider":"moonshot","dense":true,"listed":true,"n_measured":54,"n_estimated":77},{"id":321,"slug":"kimi-k2-5-reasoning","name":"Kimi K2.5 (Reasoning)","provider":"moonshot","dense":true,"listed":true,"n_measured":21,"n_estimated":91},{"id":318,"slug":"kimi-2-6","name":"Kimi K2.6","provider":"moonshot","dense":true,"listed":true,"n_measured":50,"n_estimated":89},{"id":322,"slug":"kimi-k2-7-code","name":"Kimi K2.7 Code","provider":"moonshot","dense":true,"listed":true,"n_measured":19,"n_estimated":80},{"id":323,"slug":"kimi-k3","name":"Kimi K3","provider":"moonshot","dense":true,"listed":true,"n_measured":57,"n_estimated":69},{"id":119,"slug":"decision-index-0-3-kodiak-v0-2-1b-accuracy","name":"Kodiak v0.2 1B (accuracy)","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":332,"slug":"lfm2-5-2-6b","name":"LFM2.5-2.6B","provider":"other","dense":true,"listed":true,"n_measured":15,"n_estimated":74},{"id":123,"slug":"decision-index-0-3-lfm2-5-2-6b-rlcd","name":"LFM2.5-2.6B-RLCD","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":333,"slug":"lfm2-5-230m","name":"LFM2.5-230M","provider":"other","dense":true,"listed":true,"n_measured":6,"n_estimated":25},{"id":124,"slug":"decision-index-0-3-lfm2-5-350m-rlcd","name":"LFM2.5-350M-RLCD","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":334,"slug":"lfm2-5-8b-a1b","name":"LFM2.5-8B-A1B","provider":"other","dense":true,"listed":true,"n_measured":14,"n_estimated":35},{"id":335,"slug":"lfm2-5-colbert-350m","name":"LFM2.5-ColBERT-350M","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":336,"slug":"lfm2-5-embedding-350m","name":"LFM2.5-Embedding-350M","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":337,"slug":"lfm2-5-vl-1-6b-extract","name":"LFM2.5-VL-1.6B-Extract","provider":"other","dense":true,"listed":true,"n_measured":9,"n_estimated":46},{"id":338,"slug":"lfm2-5-vl-3b","name":"LFM2.5-VL-3B","provider":"other","dense":true,"listed":true,"n_measured":8,"n_estimated":17},{"id":339,"slug":"lfm2-5-vl-450m","name":"LFM2.5-VL-450M","provider":"other","dense":true,"listed":true,"n_measured":6,"n_estimated":31},{"id":340,"slug":"lfm2-5-vl-450m-extract","name":"LFM2.5-VL-450M-Extract","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":350,"slug":"llada2-2-flash","name":"LLaDA2.2-flash","provider":"other","dense":true,"listed":true,"n_measured":7,"n_estimated":43},{"id":351,"slug":"llada2-2-mini","name":"LLaDA2.2-mini","provider":"other","dense":true,"listed":true,"n_measured":8,"n_estimated":46},{"id":324,"slug":"laguna-m-1","name":"Laguna M.1","provider":"other","dense":true,"listed":true,"n_measured":9,"n_estimated":41},{"id":325,"slug":"laguna-s-2-1","name":"Laguna S 2.1","provider":"other","dense":true,"listed":true,"n_measured":5,"n_estimated":33},{"id":327,"slug":"laguna-xs-2-1","name":"Laguna XS 2.1","provider":"other","dense":true,"listed":true,"n_measured":4,"n_estimated":22},{"id":326,"slug":"laguna-xs-2","name":"Laguna XS.2","provider":"other","dense":true,"listed":true,"n_measured":9,"n_estimated":41},{"id":120,"slug":"decision-index-0-3-lavoir","name":"Lavoir","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":328,"slug":"laya","name":"Laya","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":329,"slug":"laya-multilingual","name":"Laya multilingual","provider":"other","dense":false,"listed":true,"n_measured":0,"n_estimated":1},{"id":330,"slug":"laya-typed-decisions","name":"Laya typed-decisions","provider":"other","dense":false,"listed":true,"n_measured":0,"n_estimated":1},{"id":122,"slug":"decision-index-0-3-lev1-3-7b","name":"Lev1 3.7B","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":341,"slug":"ling-2-6-flash","name":"Ling 2.6 Flash","provider":"other","dense":true,"listed":true,"n_measured":13,"n_estimated":72},{"id":342,"slug":"ling-3-0-flash","name":"Ling 3.0 Flash","provider":"other","dense":true,"listed":true,"n_measured":33,"n_estimated":81},{"id":344,"slug":"ling-3-0-flash-fp8","name":"Ling 3.0 Flash FP8","provider":"other","dense":true,"listed":true,"n_measured":14,"n_estimated":74},{"id":343,"slug":"ling-3-0-flash-fin","name":"Ling 3.0 Flash Fin","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":2},{"id":345,"slug":"ling-3-0-flash-vl","name":"Ling 3.0 Flash VL","provider":"other","dense":true,"listed":true,"n_measured":9,"n_estimated":77},{"id":346,"slug":"ling-3-0-tiny","name":"Ling 3.0 Tiny","provider":"other","dense":true,"listed":true,"n_measured":8,"n_estimated":65},{"id":347,"slug":"ling-3-1-flash","name":"Ling 3.1 Flash","provider":"other","dense":true,"listed":true,"n_measured":13,"n_estimated":67},{"id":349,"slug":"litjev","name":"LitJev Qwen3.8-27B","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":352,"slug":"llama-3-1-405b","name":"Llama 3.1 405B","provider":"other","dense":true,"listed":true,"n_measured":8,"n_estimated":34},{"id":353,"slug":"llama-4-maverick","name":"Llama 4 Maverick","provider":"other","dense":true,"listed":true,"n_measured":14,"n_estimated":91},{"id":354,"slug":"llama-4-scout","name":"Llama 4 Scout","provider":"other","dense":true,"listed":true,"n_measured":14,"n_estimated":91},{"id":356,"slug":"longcat-flash-lite-sparse","name":"LongCat-Flash-Lite-Sparse","provider":"other","dense":true,"listed":true,"n_measured":15,"n_estimated":74},{"id":357,"slug":"mai-code-1-1-flash","name":"MAI-Code-1.1-Flash","provider":"other","dense":false,"listed":true,"n_measured":2,"n_estimated":16},{"id":358,"slug":"mai-thinking-1","name":"MAI-Thinking-1","provider":"other","dense":true,"listed":true,"n_measured":11,"n_estimated":47},{"id":359,"slug":"malkuth-2b","name":"Malkuth-2B","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":360,"slug":"malkuth-4b","name":"Malkuth-4B","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":361,"slug":"manchego-v2-1","name":"Manchego v2.1","provider":"other","dense":false,"listed":true,"n_measured":0,"n_estimated":1},{"id":362,"slug":"mellum2-12b-a2-5b-instruct","name":"Mellum2-12B-A2.5B-Instruct","provider":"other","dense":true,"listed":true,"n_measured":6,"n_estimated":32},{"id":363,"slug":"mellum2-12b-a2-5b-thinking","name":"Mellum2-12B-A2.5B-Thinking","provider":"other","dense":true,"listed":true,"n_measured":6,"n_estimated":32},{"id":364,"slug":"mercury-2-5","name":"Mercury 2.5","provider":"other","dense":true,"listed":true,"n_measured":13,"n_estimated":76},{"id":371,"slug":"mimo-v2-flash","name":"MiMo-V2-Flash","provider":"xiaomi","dense":true,"listed":true,"n_measured":15,"n_estimated":72},{"id":372,"slug":"mimo-v2-omni","name":"MiMo-V2-Omni","provider":"xiaomi","dense":true,"listed":true,"n_measured":11,"n_estimated":64},{"id":373,"slug":"mimo-v2-pro","name":"MiMo-V2-Pro","provider":"xiaomi","dense":true,"listed":true,"n_measured":12,"n_estimated":58},{"id":367,"slug":"mimo-v2-5","name":"MiMo-V2.5","provider":"xiaomi","dense":true,"listed":true,"n_measured":13,"n_estimated":67},{"id":368,"slug":"mimo-v2-5-pro","name":"MiMo-V2.5-Pro","provider":"xiaomi","dense":true,"listed":true,"n_measured":25,"n_estimated":77},{"id":369,"slug":"mimo-v2-6-flash","name":"MiMo-V2.6-Flash","provider":"xiaomi","dense":true,"listed":true,"n_measured":21,"n_estimated":82},{"id":370,"slug":"mimo-v2-6-pro","name":"MiMo-V2.6-Pro","provider":"xiaomi","dense":true,"listed":true,"n_measured":23,"n_estimated":68},{"id":374,"slug":"minicpm5-1b","name":"MiniCPM5-1B","provider":"other","dense":true,"listed":true,"n_measured":13,"n_estimated":32},{"id":375,"slug":"minicpm5-2b","name":"MiniCPM5-2B","provider":"other","dense":true,"listed":true,"n_measured":26,"n_estimated":73},{"id":376,"slug":"minimax-m2-5","name":"MiniMax M2.5","provider":"minimax","dense":false,"listed":true,"n_measured":1,"n_estimated":8},{"id":377,"slug":"minimax-m2-7","name":"MiniMax M2.7","provider":"minimax","dense":true,"listed":true,"n_measured":30,"n_estimated":75},{"id":378,"slug":"minimax-m3","name":"MiniMax M3","provider":"minimax","dense":true,"listed":true,"n_measured":42,"n_estimated":78},{"id":379,"slug":"mirror","name":"Mirror","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":380,"slug":"mistral-large-2","name":"Mistral Large 2","provider":"mistral","dense":true,"listed":true,"n_measured":7,"n_estimated":33},{"id":381,"slug":"mistral-large-3","name":"Mistral Large 3","provider":"mistral","dense":true,"listed":true,"n_measured":13,"n_estimated":87},{"id":382,"slug":"mistral-large-4","name":"Mistral Large 4","provider":"mistral","dense":true,"listed":true,"n_measured":12,"n_estimated":85},{"id":383,"slug":"mistral-medium-3","name":"Mistral Medium 3","provider":"mistral","dense":true,"listed":true,"n_measured":9,"n_estimated":46},{"id":384,"slug":"mistral-medium-3-5-128b","name":"Mistral Medium 3.5 128B","provider":"mistral","dense":true,"listed":true,"n_measured":20,"n_estimated":91},{"id":385,"slug":"mistral-small-4","name":"Mistral Small 4","provider":"mistral","dense":true,"listed":true,"n_measured":13,"n_estimated":87},{"id":386,"slug":"mistral-small-4-reasoning","name":"Mistral Small 4 (Reasoning)","provider":"mistral","dense":true,"listed":true,"n_measured":13,"n_estimated":87},{"id":392,"slug":"mxbai-rerank-base-v2","name":"Mixedbread mxbai-rerank-base-v2","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":387,"slug":"muse-glimmer-30b","name":"Muse Glimmer 30B","provider":"meta","dense":true,"listed":true,"n_measured":31,"n_estimated":95},{"id":388,"slug":"muse-spark","name":"Muse Spark","provider":"meta","dense":true,"listed":true,"n_measured":37,"n_estimated":82},{"id":389,"slug":"muse-spark-1-1","name":"Muse Spark 1.1","provider":"meta","dense":true,"listed":true,"n_measured":31,"n_estimated":83},{"id":390,"slug":"muse-spark-1-2","name":"Muse Spark 1.2","provider":"meta","dense":true,"listed":true,"n_measured":17,"n_estimated":72},{"id":391,"slug":"muse-spark-1-3","name":"Muse Spark 1.3","provider":"meta","dense":true,"listed":true,"n_measured":26,"n_estimated":70},{"id":403,"slug":"ninfer-qwen3-8-27b","name":"NInfer Qwen3.8-27B NVFP4","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":404,"slug":"ninfer-qwen3-8-27b-t1-5","name":"NInfer Qwen3.8-27B NVFP4 (T=1.5)","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":405,"slug":"ninfer-qwen3-8-flash-next","name":"NInfer Qwen3.8-Flash-Next mixed","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":393,"slug":"needle-3","name":"Needle 3 (2-bit)","provider":"other","dense":false,"listed":true,"n_measured":0,"n_estimated":1},{"id":394,"slug":"needle-3-tools","name":"Needle 3 (options as tools)","provider":"other","dense":false,"listed":true,"n_measured":0,"n_estimated":1},{"id":396,"slug":"nemotron-3-nano-30b","name":"Nemotron 3 Nano 30B","provider":"nvidia","dense":true,"listed":true,"n_measured":12,"n_estimated":75},{"id":397,"slug":"nemotron-3-nano-omni-30b-a3b","name":"Nemotron 3 Nano Omni 30B A3B","provider":"nvidia","dense":true,"listed":true,"n_measured":21,"n_estimated":87},{"id":398,"slug":"nemotron-3-super-100b","name":"Nemotron 3 Super 100B","provider":"nvidia","dense":true,"listed":true,"n_measured":14,"n_estimated":75},{"id":399,"slug":"nemotron-3-ultra","name":"Nemotron 3 Ultra","provider":"nvidia","dense":true,"listed":true,"n_measured":41,"n_estimated":66},{"id":395,"slug":"nemotron-3-5-lightning-30b-a3b-nvfp4","name":"Nemotron 3.5 Lightning 30B A3B NVFP4","provider":"nvidia","dense":true,"listed":true,"n_measured":21,"n_estimated":80},{"id":400,"slug":"nemotron-diffusion-8b","name":"Nemotron Diffusion 8B (optimized vLLM)","provider":"nvidia","dense":false,"listed":true,"n_measured":0,"n_estimated":1},{"id":401,"slug":"nemotron-ultra-253b","name":"Nemotron Ultra 253B","provider":"nvidia","dense":true,"listed":true,"n_measured":7,"n_estimated":33},{"id":127,"slug":"decision-index-0-3-neohorse-jev-4b","name":"NeoHorse-Jev-4B","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":406,"slug":"north-micro-vision-instruct","name":"North Micro Vision Instruct","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":6},{"id":407,"slug":"north-mini-code-1-0","name":"North Mini Code","provider":"other","dense":true,"listed":true,"n_measured":10,"n_estimated":69},{"id":408,"slug":"north-small-translate-1-0","name":"North Small Translate","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":409,"slug":"nova-pro","name":"Nova Pro","provider":"other","dense":true,"listed":true,"n_measured":9,"n_estimated":46},{"id":128,"slug":"decision-index-0-3-onejev-0-8b","name":"OneJev 0.8B","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":419,"slug":"open-jev-json-canvas-joshuasp","name":"Open Jev JSON Canvas","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":420,"slug":"open-jev-zefan-2b","name":"Open-Jev 2B","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":421,"slug":"open-jev-zefan-9b","name":"Open-Jev 9B","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":422,"slug":"opendecision","name":"OpenDecision ModernBERT-large","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":425,"slug":"openjev-thinking","name":"OpenJev (thinking, BF16)","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":423,"slug":"openjev-razorback16","name":"OpenJev DiffusionGemma 26B-A4B NVFP4","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":428,"slug":"opensourcejev-qwen35-4b-q4km","name":"OpenSourceJev Qwen3.5-4B Q4_K_M","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":429,"slug":"ornith-1-0-35b","name":"Ornith-1.0-35B","provider":"other","dense":true,"listed":true,"n_measured":6,"n_estimated":35},{"id":430,"slug":"ornith-1-0-397b","name":"Ornith-1.0-397B","provider":"other","dense":true,"listed":true,"n_measured":6,"n_estimated":35},{"id":431,"slug":"ornith-1-0-9b","name":"Ornith-1.0-9B","provider":"other","dense":true,"listed":true,"n_measured":6,"n_estimated":35},{"id":432,"slug":"ornith-1-5-35b-a3b","name":"Ornith-1.5-35B-A3B","provider":"other","dense":true,"listed":true,"n_measured":17,"n_estimated":71},{"id":433,"slug":"ornith-1-5-397b","name":"Ornith-1.5-397B","provider":"other","dense":true,"listed":true,"n_measured":17,"n_estimated":71},{"id":434,"slug":"ornith-1-5-9b","name":"Ornith-1.5-9B","provider":"other","dense":true,"listed":true,"n_measured":15,"n_estimated":73},{"id":435,"slug":"pareto-26-10-preview","name":"Pareto 26.10 Preview","provider":"other","dense":true,"listed":true,"n_measured":3,"n_estimated":29},{"id":436,"slug":"pareto-26-9","name":"Pareto 26.9","provider":"other","dense":true,"listed":true,"n_measured":4,"n_estimated":39},{"id":441,"slug":"pplx-decider-v1-27b","name":"Perplexity Decider v1 27B","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":131,"slug":"decision-index-0-3-perplexity-decider-v1-1-27b","name":"Perplexity Decider v1.1 (27B)","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":437,"slug":"phi-4","name":"Phi-4","provider":"other","dense":true,"listed":true,"n_measured":9,"n_estimated":36},{"id":438,"slug":"phi-4-multimodal-instruct","name":"Phi-4 Multimodal Instruct","provider":"other","dense":true,"listed":true,"n_measured":4,"n_estimated":38},{"id":439,"slug":"plumb-4b","name":"Plumb-4B","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":440,"slug":"pokee-isaac-28b","name":"Pokee-Isaac 28B","provider":"other","dense":true,"listed":true,"n_measured":3,"n_estimated":12},{"id":444,"slug":"quasar-438b","name":"Quasar 438B","provider":"other","dense":true,"listed":true,"n_measured":11,"n_estimated":76},{"id":132,"slug":"decision-index-0-3-quyet-1-0-large","name":"Quyet-1.0-Large","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":457,"slug":"qwen3-6-max-preview","name":"Qwen 3.6 Max (preview)","provider":"alibaba","dense":true,"listed":true,"n_measured":17,"n_estimated":57},{"id":133,"slug":"decision-index-0-3-qwen-2-5-1b-rlcd","name":"Qwen-2.5-1B-RLCD","provider":"alibaba","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":445,"slug":"qwen2-5-coder-32b-instruct","name":"Qwen2.5 Coder 32B Instruct","provider":"alibaba","dense":true,"listed":true,"n_measured":3,"n_estimated":26},{"id":446,"slug":"qwen3-235b-2507","name":"Qwen3 235B 2507","provider":"alibaba","dense":true,"listed":true,"n_measured":4,"n_estimated":20},{"id":447,"slug":"qwen3-235b-2507-reasoning","name":"Qwen3 235B 2507 (Reasoning)","provider":"alibaba","dense":false,"listed":true,"n_measured":2,"n_estimated":4},{"id":467,"slug":"qwen3-max","name":"Qwen3 Max","provider":"alibaba","dense":true,"listed":true,"n_measured":10,"n_estimated":53},{"id":468,"slug":"qwen3-omni-30b-a3b-instruct","name":"Qwen3-Omni-30B-A3B-Instruct","provider":"alibaba","dense":true,"listed":true,"n_measured":9,"n_estimated":46},{"id":469,"slug":"qwen3-omni-30b-a3b-thinking","name":"Qwen3-Omni-30B-A3B-Thinking","provider":"alibaba","dense":true,"listed":true,"n_measured":9,"n_estimated":46},{"id":470,"slug":"qwen3-reranker-4b","name":"Qwen3-Reranker-4B","provider":"alibaba","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":451,"slug":"qwen3-5-397b","name":"Qwen3.5 397B","provider":"alibaba","dense":true,"listed":true,"n_measured":46,"n_estimated":81},{"id":452,"slug":"qwen3-5-397b-reasoning","name":"Qwen3.5 397B (Reasoning)","provider":"alibaba","dense":true,"listed":true,"n_measured":9,"n_estimated":46},{"id":453,"slug":"qwen3-5-flash","name":"Qwen3.5 Flash","provider":"alibaba","dense":true,"listed":true,"n_measured":6,"n_estimated":36},{"id":454,"slug":"qwen3-5-plus","name":"Qwen3.5 Plus","provider":"alibaba","dense":true,"listed":true,"n_measured":4,"n_estimated":28},{"id":366,"slug":"mghafiri-qwen3-5-0-8b-decision-model","name":"Qwen3.5-0.8B Decision Model","provider":"alibaba","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":448,"slug":"qwen3-5-122b-a10b","name":"Qwen3.5-122B-A10B","provider":"alibaba","dense":true,"listed":true,"n_measured":28,"n_estimated":90},{"id":449,"slug":"qwen3-5-27b","name":"Qwen3.5-27B","provider":"alibaba","dense":true,"listed":true,"n_measured":25,"n_estimated":79},{"id":450,"slug":"qwen3-5-35b-a3b","name":"Qwen3.5-35B-A3B","provider":"alibaba","dense":true,"listed":true,"n_measured":25,"n_estimated":79},{"id":471,"slug":"qwen35-9b-jev-data-mix-v2","name":"Qwen3.5-9B Jev-like data-mix v2","provider":"alibaba","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":458,"slug":"qwen3-6-plus","name":"Qwen3.6 Plus","provider":"alibaba","dense":true,"listed":true,"n_measured":58,"n_estimated":78},{"id":455,"slug":"qwen3-6-27b","name":"Qwen3.6-27B","provider":"alibaba","dense":true,"listed":true,"n_measured":44,"n_estimated":85},{"id":456,"slug":"qwen3-6-35b-a3b","name":"Qwen3.6-35B-A3B","provider":"alibaba","dense":true,"listed":true,"n_measured":47,"n_estimated":83},{"id":459,"slug":"qwen3-7-max","name":"Qwen3.7 Max","provider":"alibaba","dense":true,"listed":true,"n_measured":46,"n_estimated":66},{"id":460,"slug":"qwen3-7-plus","name":"Qwen3.7 Plus","provider":"alibaba","dense":true,"listed":true,"n_measured":54,"n_estimated":78},{"id":464,"slug":"qwen3-8-max","name":"Qwen3.8 Max","provider":"alibaba","dense":true,"listed":true,"n_measured":43,"n_estimated":65},{"id":465,"slug":"qwen3-8-max-preview","name":"Qwen3.8 Max Preview","provider":"alibaba","dense":true,"listed":true,"n_measured":13,"n_estimated":83},{"id":461,"slug":"qwen3-8-27b","name":"Qwen3.8-27B","provider":"alibaba","dense":true,"listed":true,"n_measured":44,"n_estimated":79},{"id":462,"slug":"qwen3-8-27b-chutes","name":"Qwen3.8-27B (Chutes TEE)","provider":"alibaba","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":463,"slug":"qwen3-8-flash-next","name":"Qwen3.8-Flash-Next","provider":"alibaba","dense":true,"listed":true,"n_measured":30,"n_estimated":85},{"id":466,"slug":"qwen3-8-omni-flash","name":"Qwen3.8-Omni-Flash","provider":"alibaba","dense":true,"listed":true,"n_measured":15,"n_estimated":53},{"id":135,"slug":"decision-index-0-3-rsi-jev-v5-0-vl-3b","name":"RSI-Jev v5.0 VL 3B","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":472,"slug":"raw-phi-4-mini","name":"Raw Phi-4 mini direct logits","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":473,"slug":"raw-qwen3-0-6b","name":"Raw Qwen3 0.6B direct logits","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":474,"slug":"raw-qwen3-1-7b","name":"Raw Qwen3 1.7B direct logits","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":475,"slug":"raw-qwen3-4b-instruct-2507","name":"Raw Qwen3 4B Instruct 2507 direct logits","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":476,"slug":"raw-qwen3-8b","name":"Raw Qwen3 8B direct logits","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":136,"slug":"decision-index-0-3-ryotide-qwen-v0-3-1","name":"Ryotide Qwen v0.3.1","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":503,"slug":"swe-1-7","name":"SWE-1.7","provider":"other","dense":true,"listed":true,"n_measured":3,"n_estimated":16},{"id":504,"slug":"swe-2","name":"SWE-2","provider":"other","dense":true,"listed":true,"n_measured":4,"n_estimated":14},{"id":480,"slug":"sakana-fugu","name":"Sakana Fugu","provider":"other","dense":true,"listed":true,"n_measured":9,"n_estimated":52},{"id":482,"slug":"sakana-fugu-ultra","name":"Sakana Fugu-Ultra","provider":"other","dense":true,"listed":true,"n_measured":9,"n_estimated":52},{"id":483,"slug":"sarvam-105b","name":"Sarvam 105B","provider":"other","dense":true,"listed":true,"n_measured":8,"n_estimated":34},{"id":484,"slug":"sarvam-30b","name":"Sarvam 30B","provider":"other","dense":true,"listed":true,"n_measured":8,"n_estimated":34},{"id":485,"slug":"seed-2-1-pro","name":"Seed 2.1 Pro","provider":"other","dense":true,"listed":true,"n_measured":9,"n_estimated":35},{"id":486,"slug":"seed-2-1-turbo","name":"Seed 2.1 Turbo","provider":"other","dense":true,"listed":true,"n_measured":9,"n_estimated":35},{"id":137,"slug":"decision-index-0-3-semif","name":"SemIf","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":487,"slug":"semif-qwen3-5-4b","name":"SemIf Qwen3.5-4B","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":138,"slug":"decision-index-0-3-sieve-2b","name":"Sieve 2B","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":139,"slug":"decision-index-0-3-sieve-9b","name":"Sieve 9B","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":140,"slug":"decision-index-0-3-sifr-0-8b-v3-1","name":"Sifr 0.8B v3.1","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":488,"slug":"simplejev-qwen3-5-0-8b","name":"SimpleJev Qwen3.5-0.8B","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":489,"slug":"simplejev-qwen3-6-35b-a3b","name":"SimpleJev Qwen3.6-35B-A3B","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":490,"slug":"simplejev-qwen3-8-27b","name":"SimpleJev Qwen3.8-27B","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":492,"slug":"solar-open-2","name":"Solar Open 2","provider":"other","dense":true,"listed":true,"n_measured":11,"n_estimated":59},{"id":493,"slug":"solar-pro-2","name":"Solar Pro 2","provider":"other","dense":true,"listed":true,"n_measured":8,"n_estimated":34},{"id":494,"slug":"solar-pro-3","name":"Solar Pro 3","provider":"other","dense":true,"listed":true,"n_measured":11,"n_estimated":68},{"id":495,"slug":"solar-pro-4","name":"Solar Pro 4","provider":"other","dense":true,"listed":true,"n_measured":17,"n_estimated":82},{"id":141,"slug":"decision-index-0-3-solomon-v1-1","name":"Solomon v1.1","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":496,"slug":"soofi-s-30b-a3b","name":"Soofi S 30B-A3B","provider":"other","dense":true,"listed":true,"n_measured":3,"n_estimated":21},{"id":498,"slug":"standardone-8b","name":"Standard One 8B","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":499,"slug":"step-3-7-flash","name":"Step 3.7 Flash","provider":"stepfun","dense":true,"listed":true,"n_measured":23,"n_estimated":92},{"id":500,"slug":"step-5-preview","name":"Step 5 Preview","provider":"stepfun","dense":true,"listed":true,"n_measured":34,"n_estimated":79},{"id":161,"slug":"decision-index-rune-26b-a4b-v3","name":"Surogate Rune 26B-A4B v3","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":501,"slug":"surogate-rune-26b-a4b-v3-rtxpro6000-a2","name":"Surogate Rune 26B-A4B v3 (RTX PRO 6000)","provider":"other","dense":false,"listed":true,"n_measured":0,"n_estimated":1},{"id":507,"slug":"ternary-bonsai-2-27b","name":"Ternary Bonsai 2 27B","provider":"other","dense":true,"listed":true,"n_measured":14,"n_estimated":54},{"id":143,"slug":"decision-index-0-3-tev1-0-8b-experimental","name":"Tev1-0.8B-experimental","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":144,"slug":"decision-index-0-3-tev1-4b-experimental","name":"Tev1-4B-experimental","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":146,"slug":"decision-index-0-3-torchcast-decision-27b","name":"Torchcast Decision 27B","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":508,"slug":"trinity-large-preview","name":"Trinity-Large-Preview","provider":"other","dense":true,"listed":true,"n_measured":15,"n_estimated":74},{"id":509,"slug":"trinity-large-thinking","name":"Trinity-Large-Thinking","provider":"other","dense":true,"listed":true,"n_measured":15,"n_estimated":77},{"id":511,"slug":"ui-mate-27b","name":"UI-Mate-27B","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":12},{"id":512,"slug":"ui-mate-9b","name":"UI-Mate-9B","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":12},{"id":513,"slug":"ultravox-v0-6-llama-3-3-70b","name":"Ultravox v0.6 Llama 3.3 70B","provider":"other","dense":true,"listed":true,"n_measured":10,"n_estimated":74},{"id":147,"slug":"decision-index-0-3-verdict","name":"Verdict","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":154,"slug":"decision-index-0-3-von-1-3","name":"Von 1.3","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":515,"slug":"von-395m","name":"Von 395M","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":516,"slug":"winnow-12b","name":"Winnow-12B Q8","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":155,"slug":"decision-index-0-3-winnow-e4b","name":"Winnow-E4B","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":156,"slug":"decision-index-0-3-xor-1-2","name":"Xor 1.2","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":157,"slug":"decision-index-0-3-xor-26b-a4b","name":"Xor 26B-A4B","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":517,"slug":"zaya1-74b-preview","name":"ZAYA1-74B-Preview","provider":"other","dense":true,"listed":true,"n_measured":6,"n_estimated":37},{"id":518,"slug":"zaya1-8b","name":"ZAYA1-8B","provider":"other","dense":true,"listed":true,"n_measured":10,"n_estimated":36},{"id":519,"slug":"zaya1-vl-8b","name":"ZAYA1-VL-8B","provider":"other","dense":false,"listed":true,"n_measured":3,"n_estimated":6},{"id":520,"slug":"zerank-2","name":"ZeroEntropy zerank-2","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":158,"slug":"decision-index-0-3-zev","name":"Zev","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":17,"slug":"classifier-dev-fast","name":"classifier.dev (fast)","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":348,"slug":"liquid-d1","name":"d1","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":54,"slug":"d1-3b","name":"d1-3B","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":55,"slug":"d1-omni-600m","name":"d1-omni-600M","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":56,"slug":"decider-2b","name":"decider-2b","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":57,"slug":"decider-35b-a3b","name":"decider-35b-a3b","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":58,"slug":"decider-4b-v2","name":"decider-4b v2","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":74,"slug":"decision-index-0-3-decisio-0-4-0-gemma-4-12b","name":"decisio 0.4.0 (Gemma 4 12B)","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":75,"slug":"decision-index-0-3-decisio-0-4-0-qwen3-6-35b-a3b","name":"decisio 0.4.0 (Qwen3.6-35B-A3B)","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":163,"slug":"decision-machine-1","name":"decision-machine-1","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":76,"slug":"decision-index-0-3-deck31b-gemma-4-31b","name":"deck31b (Gemma 4 31B)","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":176,"slug":"djev","name":"djev","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":177,"slug":"djev-thinking","name":"djev (thinking)","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":178,"slug":"dots3-note-preview","name":"dots3-note Preview","provider":"other","dense":true,"listed":true,"n_measured":27,"n_estimated":82},{"id":80,"slug":"decision-index-0-3-ezjev-4b-s2","name":"ezjev 4B s2","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":299,"slug":"jeff","name":"jeff","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":301,"slug":"jev-local","name":"jev-local Qwen3.5-9B","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":105,"slug":"decision-index-0-3-jiwo-0-8b","name":"jiwo 0.8B","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":106,"slug":"decision-index-0-3-jiwo-4b","name":"jiwo 4B","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":309,"slug":"jqv","name":"jqv Qwen3-32B","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":314,"slug":"kev-0-5b","name":"kev 0.5B","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":315,"slug":"kev-0-6b","name":"kev 0.6B","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":316,"slug":"kev-4b","name":"kev 4B","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":317,"slug":"kev-8b","name":"kev 8B","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":121,"slug":"decision-index-0-3-lev","name":"lev","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":331,"slug":"lev-350m","name":"lev-350m","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":355,"slug":"localjev-qwen3-5-4b","name":"local-jev Qwen3.5-4B","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":365,"slug":"metask-jev-4b","name":"metask-jev-4b","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":125,"slug":"decision-index-0-3-mini-jev","name":"mini-jev","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":126,"slug":"decision-index-0-3-mmastrac-diffusiongemma-vllm-pr-57250","name":"mmastrac diffusiongemma (vLLM PR 57250)","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":410,"slug":"o1","name":"o1","provider":"openai","dense":true,"listed":true,"n_measured":13,"n_estimated":55},{"id":411,"slug":"o1-preview","name":"o1-preview","provider":"openai","dense":true,"listed":true,"n_measured":3,"n_estimated":46},{"id":412,"slug":"o1-pro","name":"o1-pro","provider":"openai","dense":false,"listed":true,"n_measured":2,"n_estimated":25},{"id":413,"slug":"o3","name":"o3","provider":"openai","dense":true,"listed":true,"n_measured":11,"n_estimated":50},{"id":414,"slug":"o3-mini","name":"o3-mini","provider":"openai","dense":true,"listed":true,"n_measured":8,"n_estimated":45},{"id":415,"slug":"o3-pro","name":"o3-pro","provider":"openai","dense":false,"listed":true,"n_measured":2,"n_estimated":26},{"id":416,"slug":"o4-mini-high","name":"o4-mini (high)","provider":"openai","dense":false,"listed":true,"n_measured":2,"n_estimated":4},{"id":417,"slug":"open-alternative-jev","name":"open-alternative-jev Qwen3.5-4B","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":129,"slug":"decision-index-0-3-open-jev-pngwn","name":"open-jev (pngwn)","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":418,"slug":"open-jev-deberta-v3-large","name":"open-jev-deberta-v3-large","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":426,"slug":"openjev-verdict","name":"openJev Verdict","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":427,"slug":"openjev-verdict-1-4","name":"openJev Verdict 1.4","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":424,"slug":"openjev-sglang","name":"openjev-sglang Qwen3.6-35B-A3B","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":130,"slug":"decision-index-0-3-openvons","name":"openvons","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":442,"slug":"pplx-embed-v2-late-0-6b","name":"pplx-embed-v2-late-0.6b","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":443,"slug":"pplx-embed-v2-late-9b","name":"pplx-embed-v2-late-9b","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":134,"slug":"decision-index-0-3-razorback16-openjev-diffusiongemma-nvfp4-vllm","name":"razorback16 openjev diffusiongemma (NVFP4, vLLM)","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":479,"slug":"reflex-4b","name":"reflex 4B","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":478,"slug":"reflex-27b","name":"reflex-27b","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":162,"slug":"decision-index-simple-jev-qwen3-8-27b","name":"simple-jev · Qwen3.8-27B","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":491,"slug":"smalljev","name":"smalljev semantic-v9","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":497,"slug":"spark-s1-4b-v6","name":"spark-s1-4b-v6","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":502,"slug":"swanone","name":"swanOne","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":506,"slug":"system-one-sg","name":"system-one Qwen3-8B","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":142,"slug":"decision-index-0-3-system-one-gemma","name":"system-one-gemma","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":505,"slug":"system-one-open","name":"system-one-open","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":145,"slug":"decision-index-0-3-this-that-1-2","name":"this-that 1.2","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":510,"slug":"typecastlm","name":"typecastlm","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0},{"id":148,"slug":"decision-index-0-3-vllm-sr-decision-2-0-eos-0-8b","name":"vLLM-SR Decision 2.0 Eos 0.8B","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":149,"slug":"decision-index-0-3-vllm-sr-decision-2-0-kai-0-6b","name":"vLLM-SR Decision 2.0 Kai 0.6B","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":150,"slug":"decision-index-0-3-vllm-sr-decision-2-0-lux-9b","name":"vLLM-SR Decision 2.0 Lux 9B","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":151,"slug":"decision-index-0-3-vllm-sr-decision-2-0-nox-4b","name":"vLLM-SR Decision 2.0 Nox 4B","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":152,"slug":"decision-index-0-3-vllm-sr-decision-2-0-sol-2b","name":"vLLM-SR Decision 2.0 Sol 2B","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":153,"slug":"decision-index-0-3-vllm-sr-decision-2-0-vega-27b","name":"vLLM-SR Decision 2.0 Vega 27B","provider":"other","dense":false,"listed":false,"n_measured":0,"n_estimated":0},{"id":514,"slug":"verdict-small","name":"verdict-small","provider":"other","dense":false,"listed":true,"n_measured":1,"n_estimated":0}]}