Long context
MRCRv2 leaderboard
As of 2026-10-10, the highest measured score on MRCRv2 is 93.6% by Sakana Fugu-Ultra. 177 more models have estimated scores, calibrated from the benchmarks they were measured on.
Measured scores: benchlm.ai.
| # | Model | Score | Source |
|---|---|---|---|
| 1 | Sakana Fugu-Ultra | 93.6% | measured |
| 2 | Qwen3.8 Max | 92.9% | measured |
| 3 | Qwen3.7 Plus | 91.7% | measured |
| 4 | Qwen3.7 Max | 90.4% | measured |
| 5 | Sakana Fugu | 86.6% | measured |
| 6 | Kimi K3 | 86.0% | estimated ± 11.3 pp, low confidence |
| 7 | Step 5 Preview | 86.0% | estimated ± 11.3 pp, low confidence |
| 8 | MiMo-V2.6-Pro | 86.0% | estimated ± 11.3 pp, low confidence |
| 9 | Claude Fable 5.1 | 86.0% | estimated ± 11.3 pp, low confidence |
| 10 | Claude Opus 5.5 | 86.0% | estimated ± 11.3 pp, low confidence |
| 11 | GPT-5.5 | 86.0% | estimated ± 11.3 pp, low confidence |
| 12 | DeepSeek V4.1 Flash | 86.0% | estimated ± 11.3 pp, low confidence |
| 13 | GPT-5.6 Sol | 86.0% | estimated ± 11.3 pp, low confidence |
| 14 | GPT-5.6 Luna | 85.9% | estimated ± 11.3 pp, low confidence |
| 15 | GPT-6 Sol | 85.9% | estimated ± 11.3 pp, low confidence |
| 16 | GPT-5.3 Codex | 85.9% | estimated ± 11.3 pp, low confidence |
| 17 | GPT-6 Luna | 85.9% | estimated ± 11.3 pp, low confidence |
| 18 | Muse Glimmer 30B | 85.9% | estimated ± 11.3 pp, low confidence |
| 19 | GPT-5.6 Terra | 85.9% | estimated ± 11.3 pp, low confidence |
| 20 | GPT-6.1 Sol | 85.9% | estimated ± 11.3 pp, low confidence |
| 21 | Ling 3.1 Flash | 85.9% | estimated ± 11.3 pp, low confidence |
| 22 | MiniMax M3 | 85.9% | estimated ± 11.3 pp, low confidence |
| 23 | Muse Spark 1.3 | 85.9% | estimated ± 11.3 pp, low confidence |
| 24 | Claude Haiku 5.5 | 85.9% | estimated ± 11.3 pp, low confidence |
| 25 | Claude Sonnet 5.5 | 85.9% | estimated ± 11.3 pp, low confidence |
| 26 | GPT-5.2 | 85.9% | estimated ± 11.3 pp, low confidence |
| 27 | Claude Fable 5 | 85.9% | estimated ± 11.3 pp, low confidence |
| 28 | GPT-5.2-Codex | 85.9% | estimated ± 11.3 pp, low confidence |
| 29 | Claude Sonnet 5 | 85.9% | estimated ± 11.3 pp, low confidence |
| 30 | Gemini 3.1 Pro | 85.9% | estimated ± 11.3 pp, low confidence |
| 31 | GPT-5.4 | 85.9% | estimated ± 11.3 pp, low confidence |
| 32 | Qwen3.8-27B | 85.9% | estimated ± 11.3 pp, low confidence |
| 33 | Gemini 3.7 Flash | 85.9% | estimated ± 11.3 pp, low confidence |
| 34 | Gemini 3.8 Flash | 85.9% | estimated ± 11.3 pp, low confidence |
| 35 | Mistral Large 4 | 85.9% | estimated ± 11.3 pp, low confidence |
| 36 | Kimi K2.6 | 85.8% | estimated ± 11.3 pp, low confidence |
| 37 | GPT-6 Astra | 85.8% | estimated ± 11.3 pp, low confidence |
| 38 | Qwen 3.6 Max (preview) | 85.8% | estimated ± 11.3 pp, low confidence |
| 39 | DeepSeek V4 Pro 0813 | 85.8% | estimated ± 11.3 pp, low confidence |
| 40 | Grok 4.6 | 85.8% | estimated ± 11.3 pp, low confidence |
| 41 | Qwen3.8 Max Preview | 85.8% | estimated ± 11.3 pp, low confidence |
| 42 | Gemini 3.6 Flash | 85.8% | estimated ± 11.3 pp, low confidence |
| 43 | GLM-5.3-Flash | 85.8% | estimated ± 11.3 pp, low confidence |
| 44 | GPT-5.1 | 85.8% | estimated ± 11.3 pp, low confidence |
| 45 | DeepSeek V4 Flash 0731 | 85.7% | estimated ± 11.3 pp, low confidence |
| 46 | Gemini 4 Argon | 85.7% | estimated ± 11.3 pp, low confidence |
| 47 | GLM-5.3 | 85.7% | estimated ± 11.3 pp, low confidence |
| 48 | MiMo-V2.5-Pro | 85.7% | estimated ± 11.3 pp, low confidence |
| 49 | Qwen3.8-Flash-Next | 85.7% | estimated ± 11.3 pp, low confidence |
| 50 | Apodex 1.1 | 85.7% | estimated ± 11.3 pp, low confidence |
| 51 | Apodex 1.1 Mini | 85.7% | estimated ± 11.3 pp, low confidence |
| 52 | Claude Opus 5 | 85.7% | estimated ± 11.3 pp, low confidence |
| 53 | Grok 4.5 | 85.7% | estimated ± 11.3 pp, low confidence |
| 54 | Kimi K2.7 Code | 85.7% | estimated ± 11.3 pp, low confidence |
| 55 | Hy3 | 85.6% | estimated ± 11.3 pp, low confidence |
| 56 | Muse Spark 1.2 | 85.6% | estimated ± 11.3 pp, low confidence |
| 57 | Claude Opus 4.7 (Adaptive) | 85.6% | estimated ± 11.3 pp, low confidence |
| 58 | GLM-5.2 | 85.5% | estimated ± 11.3 pp, low confidence |
| 59 | Ling 3.0 Flash VL | 85.5% | estimated ± 11.3 pp, low confidence |
| 60 | MiniMax M2.7 | 85.5% | estimated ± 11.3 pp, low confidence |
| 61 | Qwen3.6 Plus | 85.5% | estimated ± 11.3 pp, low confidence |
| 62 | GPT-5 (high) | 85.5% | estimated ± 11.3 pp, low confidence |
| 63 | Claude Opus 4.6 (Adaptive) | 85.5% | estimated ± 11.3 pp, low confidence |
| 64 | Kimi K2.5 | 85.5% | estimated ± 11.3 pp, low confidence |
| 65 | Kimi K2.5 (Reasoning) | 85.5% | estimated ± 11.3 pp, low confidence |
| 66 | Muse Spark | 85.5% | estimated ± 11.3 pp, low confidence |
| 67 | Claude Opus 4.8 | 85.4% | estimated ± 11.3 pp, low confidence |
| 68 | Muse Spark 1.1 | 85.4% | estimated ± 11.3 pp, low confidence |
| 69 | Qwen3.5-27B | 85.4% | estimated ± 11.3 pp, low confidence |
| 70 | Claude Opus 4.5 Thinking | 85.3% | estimated ± 11.3 pp, low confidence |
| 71 | Inkling | 85.3% | estimated ± 11.3 pp, low confidence |
| 72 | Qwen3.6-27B | 85.3% | estimated ± 11.3 pp, low confidence |
| 73 | GPT-5.4 mini | 85.2% | estimated ± 11.3 pp, low confidence |
| 74 | Ternary Bonsai 2 27B | 85.2% | estimated ± 11.3 pp, low confidence |
| 75 | GPT-5.4 nano | 85.1% | estimated ± 11.3 pp, low confidence |
| 76 | Grok 4.7 | 85.1% | estimated ± 11.3 pp, low confidence |
| 77 | Quasar 438B | 84.9% | estimated ± 11.3 pp, low confidence |
| 78 | Qwen3.5-122B-A10B | 84.9% | estimated ± 11.3 pp, low confidence |
| 79 | Claude 4.1 Opus Thinking | 84.8% | estimated ± 11.3 pp, low confidence |
| 80 | Gemini 3 Pro | 84.8% | estimated ± 11.3 pp, low confidence |
| 81 | GPT-5 (medium) | 84.8% | estimated ± 11.3 pp, low confidence |
| 82 | Claude Opus 4.7 | 84.6% | estimated ± 11.3 pp, low confidence |
| 83 | GLM-5 | 84.6% | estimated ± 11.3 pp, low confidence |
| 84 | Inkling-Small | 84.6% | estimated ± 11.3 pp, low confidence |
| 85 | MiMo-V2-Omni | 84.2% | estimated ± 11.3 pp, low confidence |
| 86 | o3 | 83.9% | estimated ± 11.3 pp, low confidence |
| 87 | MiMo-V2.6-Flash | 83.6% | estimated ± 11.3 pp, low confidence |
| 88 | Grok 4.1 Fast (Reasoning) | 83.3% | estimated ± 11.3 pp, low confidence |
| 89 | GLM-5.1 | 82.9% | estimated ± 11.3 pp, low confidence |
| 90 | Grok 4 Fast (Reasoning) | 82.9% | estimated ± 11.3 pp, low confidence |
| 91 | Step 3.7 Flash | 82.9% | estimated ± 11.3 pp, low confidence |
| 92 | Ling 3.0 Flash | 82.0% | estimated ± 11.3 pp, low confidence |
| 93 | Ling 3.0 Flash FP8 | 82.0% | estimated ± 11.3 pp, low confidence |
| 94 | Qwen3.5-35B-A3B | 80.1% | estimated ± 11.3 pp, low confidence |
| 95 | GLM-5-Turbo | 79.5% | estimated ± 11.3 pp, low confidence |
| 96 | Qwen3.6-35B-A3B | 79.5% | estimated ± 11.3 pp, low confidence |
| 97 | GLM-4.7 | 77.6% | estimated ± 11.3 pp, low confidence |
| 98 | Solar Pro 4 | 77.6% | estimated ± 11.3 pp, low confidence |
| 99 | Gemini 3.5 Flash | 77.3% | measured |
| 100 | Claude Opus 4.5 | 76.6% | estimated ± 11.3 pp, low confidence |
| 101 | GLM-5V-Turbo | 75.3% | estimated ± 11.3 pp, low confidence |
| 102 | Gemini 3.5 Flash-Lite | 72.2% | measured |
| 103 | GPT-5.1-Codex | 71.1% | estimated ± 11.3 pp, low confidence |
| 104 | GPT-5.1-Codex-Max | 71.1% | estimated ± 11.3 pp, low confidence |
| 105 | Mistral Medium 3.5 128B | 71.1% | estimated ± 11.3 pp, low confidence |
| 106 | Gemini 2.5 Pro | 69.6% | estimated ± 11.3 pp, low confidence |
| 107 | Gemma 4 31B | 66.4% | measured |
| 108 | Claude Sonnet 4.6 | 65.9% | estimated ± 11.3 pp, low confidence |
| 109 | GPT-4.1 | 65.9% | estimated ± 11.3 pp, low confidence |
| 110 | MiMo-V2-Pro | 65.9% | estimated ± 11.3 pp, low confidence |
| 111 | Grok 4 | 64.2% | estimated ± 11.3 pp, low confidence |
| 112 | Mercury 2.5 | 64.2% | estimated ± 11.3 pp, low confidence |
| 113 | Pokee-Isaac 28B | 60.7% | measured |
| 114 | Claude Opus 4.6 | 58.0% | estimated ± 11.3 pp, low confidence |
| 115 | Nemotron 3 Ultra | 58.0% | estimated ± 11.3 pp, low confidence |
| 116 | Hy3 Preview | 56.1% | estimated ± 11.3 pp, low confidence |
| 117 | A.X K2 | 51.6% | estimated ± 11.3 pp, low confidence |
| 118 | Nemotron 3 Super 100B | 49.7% | estimated ± 11.3 pp, low confidence |
| 119 | o1 | 45.4% | estimated ± 11.3 pp, low confidence |
| 120 | Gemma 4 26B A4B | 44.1% | measured |
| 121 | Gemma 4 12B | 43.4% | measured |
| 122 | Grok 4.3 | 41.4% | estimated ± 11.3 pp, low confidence |
| 123 | Qwen3.5 397B | 41.4% | estimated ± 11.3 pp, low confidence |
| 124 | Qwen3.5 397B (Reasoning) | 41.4% | estimated ± 11.3 pp, low confidence |
| 125 | Solar Open 2 | 32.6% | estimated ± 11.3 pp, low confidence |
| 126 | K-Exaone | 29.7% | estimated ± 11.3 pp, low confidence |
| 127 | Ling 3.0 Tiny | 27.5% | estimated ± 11.3 pp, low confidence |
| 128 | MiniCPM5-2B | 25.7% | estimated ± 11.3 pp, low confidence |
| 129 | Gemma 4 E4B | 25.4% | measured |
| 130 | DeepSeek V3.1 (Reasoning) | 24.0% | estimated ± 11.3 pp, low confidence |
| 131 | K-EXAONE 2.0 | 23.8% | estimated ± 11.3 pp, low confidence |
| 132 | DeepSeek-R1 | 23.6% | estimated ± 11.3 pp, low confidence |
| 133 | Gemini 3 Flash | 23.5% | estimated ± 11.3 pp, low confidence |
| 134 | Grok Code Fast 1 | 23.1% | estimated ± 11.3 pp, low confidence |
| 135 | Kimi K2 | 23.1% | estimated ± 11.3 pp, low confidence |
| 136 | Command A+ | 23.1% | estimated ± 11.3 pp, low confidence |
| 137 | GPT-OSS 120B | 23.0% | estimated ± 11.3 pp, low confidence |
| 138 | Llama 4 Maverick | 22.9% | estimated ± 11.3 pp, low confidence |
| 139 | Qwen3 Max | 22.9% | estimated ± 11.3 pp, low confidence |
| 140 | Gemini 2.5 Flash | 22.9% | estimated ± 11.3 pp, low confidence |
| 141 | Mistral Small 4 | 22.9% | estimated ± 11.3 pp, low confidence |
| 142 | Mistral Small 4 (Reasoning) | 22.9% | estimated ± 11.3 pp, low confidence |
| 143 | GPT-4o | 22.9% | estimated ± 11.3 pp, low confidence |
| 144 | Nemotron 3.5 Lightning 30B A3B NVFP4 | 22.9% | estimated ± 11.3 pp, low confidence |
| 145 | Granite 4.2 30B | 22.9% | estimated ± 11.3 pp, low confidence |
| 146 | DeepSeek V3.1 | 22.8% | estimated ± 11.3 pp, low confidence |
| 147 | GLM-4.5-Air | 22.8% | estimated ± 11.3 pp, low confidence |
| 148 | DeepSeek V3.2 | 22.8% | estimated ± 11.3 pp, low confidence |
| 149 | Granite 4.2 8B | 22.8% | estimated ± 11.3 pp, low confidence |
| 150 | Claude 4 Sonnet | 22.8% | estimated ± 11.3 pp, low confidence |
| 151 | GPT-4.1 mini | 22.8% | estimated ± 11.3 pp, low confidence |
| 152 | DeepSeek V3 0324 | 22.8% | estimated ± 11.3 pp, low confidence |
| 153 | Nemotron 3 Nano Omni 30B A3B | 22.8% | estimated ± 11.3 pp, low confidence |
| 154 | Celeris-1 | 22.8% | estimated ± 11.3 pp, low confidence |
| 155 | Nemotron 3 Nano 30B | 22.8% | estimated ± 11.3 pp, low confidence |
| 156 | Trinity-Large-Preview | 22.8% | estimated ± 11.3 pp, low confidence |
| 157 | Trinity-Large-Thinking | 22.8% | estimated ± 11.3 pp, low confidence |
| 158 | North Mini Code | 22.8% | estimated ± 11.3 pp, low confidence |
| 159 | MiMo-V2-Flash | 22.8% | estimated ± 11.3 pp, low confidence |
| 160 | Mistral Large 3 | 22.8% | estimated ± 11.3 pp, low confidence |
| 161 | GPT-OSS 20B | 22.8% | estimated ± 11.3 pp, low confidence |
| 162 | Solar Pro 3 | 22.8% | estimated ± 11.3 pp, low confidence |
| 163 | Grok 4.1 Fast | 22.8% | estimated ± 11.3 pp, low confidence |
| 164 | Ling 2.6 Flash | 22.8% | estimated ± 11.3 pp, low confidence |
| 165 | Mistral Medium 3 | 22.8% | estimated ± 11.3 pp, low confidence |
| 166 | DeepSeek V3 | 22.8% | estimated ± 11.3 pp, low confidence |
| 167 | Claude 3 Haiku | 22.8% | estimated ± 11.3 pp, low confidence |
| 168 | Llama 4 Scout | 22.8% | estimated ± 11.3 pp, low confidence |
| 169 | GLM-4.6 | 22.8% | estimated ± 11.3 pp, low confidence |
| 170 | Llama 3.1 405B | 22.8% | estimated ± 11.3 pp, low confidence |
| 171 | Granite 4.2 3B | 22.8% | estimated ± 11.3 pp, low confidence |
| 172 | Nova Pro | 22.8% | estimated ± 11.3 pp, low confidence |
| 173 | GPT-4.1 nano | 22.8% | estimated ± 11.3 pp, low confidence |
| 174 | Ultravox v0.6 Llama 3.3 70B | 22.8% | estimated ± 11.3 pp, low confidence |
| 175 | DeepSeek R1 Distill Qwen 32B | 22.8% | estimated ± 11.3 pp, low confidence |
| 176 | Gemma 3 27B | 22.8% | estimated ± 11.3 pp, low confidence |
| 177 | Granite-4.0-H-1B | 22.8% | estimated ± 11.3 pp, low confidence |
| 178 | LFM2.5-2.6B | 22.8% | estimated ± 11.3 pp, low confidence |
| 179 | Exaone 4.0 1.2B | 22.8% | estimated ± 11.3 pp, low confidence |
| 180 | Granite-4.0-350M | 22.8% | estimated ± 11.3 pp, low confidence |
| 181 | Granite-4.0-H-350M | 22.8% | estimated ± 11.3 pp, low confidence |
| 182 | LFM2.5-8B-A1B | 22.8% | estimated ± 11.3 pp, low confidence |
| 183 | LFM2.5-VL-1.6B-Extract | 22.8% | estimated ± 11.3 pp, low confidence |
| 184 | Phi-4 | 22.8% | estimated ± 11.3 pp, low confidence |
| 185 | Qwen3-Omni-30B-A3B-Instruct | 22.8% | estimated ± 11.3 pp, low confidence |
| 186 | Qwen3-Omni-30B-A3B-Thinking | 22.8% | estimated ± 11.3 pp, low confidence |
| 187 | Sarvam 105B | 22.8% | estimated ± 11.3 pp, low confidence |
| 188 | Sarvam 30B | 22.8% | estimated ± 11.3 pp, low confidence |
| 189 | Solar Pro 2 | 22.8% | estimated ± 11.3 pp, low confidence |
| 190 | Gemma 4 E2B | 19.1% | measured |
All leaderboards
Agentic · terminal
- AA Terminal-Bench 2.1
- AA Terminal-Bench 4.0
- Terminal-Bench 2.0
- Terminal-Bench 3.0
- Terminal-Bench 4.0
- Terminal-Bench-Science 0.1
- Terminal-Bench 2.1
- Terminal-Bench 2.1 (Vals AI)
Agentic · tools
- AA Agentic Index
- AA-AnalystAgent
- AA AutomationBench
- AA EnterpriseOps-Gym
- GDP.pdf
- AA Harvey LAB v1.0
- AA ITBench
- AA Tau3 Banking
- Agents' Last Exam
- APEX-Agents
- APEX-Agents-AA
- ApprenticeBench
- AutomationBench
- BFCL v4
- BrowseComp
- Claw-Eval
- CWE-bench v1
- CyberGym
- DeepPlanning
- DeepSearchQA
- DRACO
- ExploitGym
- GDPval-AA
- Gert Labs
- HLE w/ tools
- JobBench
- MCP Atlas
- MCP-Tasks
- OSWorld-Verified
- OSWorld 2.0
- QwenClawBench
- ResearchClawBench
- SkillsBench
- τ²-bench results
- τ³-bench results
- Toolathlon
- Toolathlon-Verified
- VITA-Bench
- WideResearch
Coding
- AA Coding Index
- AA-SciCode
- Bug Hunt Bench
- CursorBench 3.1
- CursorBench 3.2
- CursorBench 4.0
- DeepSWE
- FrontierCode 1.1 Main
- FrontierCode 1.1 Extended
- FrontierSWE v2
- LiveCodeBench Pro
- LiveCodeBench v6
- NL2Repo
- OpenHarmony Bench
- PostTrainBench v1.1
- ProgramBench
- React Native Evals
- SciCode
- SWE-bench Pro
- SWE-bench Verified
- SWE-Rebench
- Vals LiveCodeBench
- Vals SWE-bench
- Vibe Code Bench
- VulcanBench v3
Math
- AIME 2024
- AIME 2025
- AIME26
- FrontierMath (legacy)
- FrontierMath v2 (Tier 4)
- FrontierMath v2 (Tiers 1-3)
- HMMT Feb 2025
- HMMT Feb 2026
- HMMT Nov 2025
- IMOAnswerBench
- MATH-500
- MMAnswerBench
Knowledge & reasoning
- AA-GPQA Diamond
- AA-HLE
- ARC-AGI-2
- ARC-AGI-1
- ARC-AGI-3
- Artificial Analysis Intelligence Index
- BBH
- BioMysteryBench (human-difficult)
- BioMysteryBench (human-solvable)
- C-Eval
- CritPt
- GPQA
- GPQA Diamond
- HealthBench (raw)
- HealthBench Hard
- HealthBench (length-adjusted)
- HealthBench Professional
- HealthBench Professional (raw)
- HLE
- HLE w/o tools
- HLE-Verified
- JevBench 1.4
- LABBench2
- MedXpertQA (Text)
- MMLU
- MMLU-Pro
- MMLU-Pro (Arcee)
- MMLU-Redux
- MMMLU
- AA-Omniscience Accuracy
- SimpleQA
- SuperGPQA
- Vals GPQA Diamond
- Vals MMLU-Pro
Instruction following
Multilingual
Vision & documents
- AA-MMMU-Pro
- AI2D_TEST
- BabyVision
- CC-OCR
- CharXiv
- CharXiv w/o tools
- CountBench
- DynaMath
- ERQA
- MathVision
- MathVision w/ Python
- MedXpertQA (MM)
- MLVU (M-Avg)
- MMMU
- MMMU-Pro
- MMMU-Pro w/ Python
- MMVU
- OfficeQA Pro
- OmniDocBench 1.5
- RealWorldQA
- RefCOCO (avg)
- ScreenSpot Pro
- SimpleVQA
- Video-MME (w/o subtitle)
- Video-MME (with subtitle)
- VideoMMMU
- V*
- We-Math
- ZeroBench