Long context
MMLongBench-Doc leaderboard
As of 2026-10-29, the highest measured score on MMLongBench-Doc is 62.8% by Nemotron 3.5 Super VL 120B A12B. 196 more models have estimated scores, calibrated from the benchmarks they were measured on.
Measured scores: benchlm.ai.
| # | Model | Score | Source |
|---|---|---|---|
| 1 | Nemotron 3.5 Super VL 120B A12B | 62.8% | measured |
| 2 | Nemotron 3.5 Super VL 120B A12B NVFP4 | 62.8% | measured |
| 3 | Kimi K3 | 61.8% | estimated ± 2.6 pp, low confidence |
| 4 | Step 5 Preview | 61.7% | estimated ± 2.6 pp, low confidence |
| 5 | Kimi K3.1 Pro | 61.7% | estimated ± 2.6 pp, low confidence |
| 6 | MiMo-V2.6-Pro | 61.7% | estimated ± 2.6 pp, low confidence |
| 7 | Claude Fable 5.1 | 61.6% | estimated ± 2.6 pp, low confidence |
| 8 | Claude Opus 5.5 | 61.6% | estimated ± 2.6 pp, low confidence |
| 9 | GPT-5.5 | 61.6% | estimated ± 2.6 pp, low confidence |
| 10 | DeepSeek V4.1 Flash 1017 | 61.6% | estimated ± 2.6 pp, low confidence |
| 11 | GPT-5.6 Sol | 61.6% | estimated ± 2.6 pp, low confidence |
| 12 | GPT-5.6 Luna | 61.6% | estimated ± 2.6 pp, low confidence |
| 13 | GPT-6 Sol | 61.6% | estimated ± 2.6 pp, low confidence |
| 14 | Muse Wonder Full | 61.6% | estimated ± 2.6 pp, low confidence |
| 15 | GPT-5.3 Codex | 61.5% | estimated ± 2.6 pp, low confidence |
| 16 | GPT-6 Luna | 61.5% | estimated ± 2.6 pp, low confidence |
| 17 | Muse Glimmer 30B | 61.5% | estimated ± 2.6 pp, low confidence |
| 18 | GPT-5.6 Terra | 61.5% | estimated ± 2.6 pp, low confidence |
| 19 | GPT-6.1 Sol | 61.5% | estimated ± 2.6 pp, low confidence |
| 20 | Ling 3.1 Flash | 61.5% | estimated ± 2.6 pp, low confidence |
| 21 | MiniMax M3 | 61.5% | estimated ± 2.6 pp, low confidence |
| 22 | Muse Spark 1.3 | 61.5% | estimated ± 2.6 pp, low confidence |
| 23 | Claude Haiku 5.5 | 61.5% | estimated ± 2.6 pp, low confidence |
| 24 | Claude Sonnet 5.5 | 61.5% | estimated ± 2.6 pp, low confidence |
| 25 | GPT-5.2 | 61.5% | estimated ± 2.6 pp, low confidence |
| 26 | GPT-6.1 Terra | 61.5% | estimated ± 2.6 pp, low confidence |
| 27 | Claude Fable 5 | 61.5% | estimated ± 2.6 pp, low confidence |
| 28 | GPT-5.2-Codex | 61.5% | estimated ± 2.6 pp, low confidence |
| 29 | Qwen3.5 397B | 61.5% | measured |
| 30 | Claude Sonnet 5 | 61.5% | estimated ± 2.6 pp, low confidence |
| 31 | Gemini 3.1 Pro | 61.5% | estimated ± 2.6 pp, low confidence |
| 32 | GPT-5.4 | 61.5% | estimated ± 2.6 pp, low confidence |
| 33 | Qwen3.8-27B | 61.5% | estimated ± 2.6 pp, low confidence |
| 34 | Gemini 3.7 Flash | 61.5% | estimated ± 2.6 pp, low confidence |
| 35 | Gemini 3.8 Flash | 61.5% | estimated ± 2.6 pp, low confidence |
| 36 | Mistral Large 4 | 61.5% | estimated ± 2.6 pp, low confidence |
| 37 | Kimi K2.6 | 61.5% | estimated ± 2.6 pp, low confidence |
| 38 | Sage-A1.5 | 61.5% | estimated ± 2.6 pp, low confidence |
| 39 | GPT-6 Astra | 61.4% | estimated ± 2.6 pp, low confidence |
| 40 | Muse Wonder | 61.4% | estimated ± 2.6 pp, low confidence |
| 41 | Qwen 3.6 Max (preview) | 61.4% | estimated ± 2.6 pp, low confidence |
| 42 | DeepSeek V4 Pro 0813 | 61.4% | estimated ± 2.6 pp, low confidence |
| 43 | Grok 4.6 | 61.4% | estimated ± 2.6 pp, low confidence |
| 44 | Qwen3.8 Max Preview | 61.4% | estimated ± 2.6 pp, low confidence |
| 45 | Gemini 3.6 Flash | 61.4% | estimated ± 2.6 pp, low confidence |
| 46 | GPT-5.1 | 61.4% | estimated ± 2.6 pp, low confidence |
| 47 | DeepSeek V4 Flash 0731 | 61.4% | estimated ± 2.6 pp, low confidence |
| 48 | Gemini 4 Argon | 61.4% | estimated ± 2.6 pp, low confidence |
| 49 | GLM-5.3 | 61.4% | estimated ± 2.6 pp, low confidence |
| 50 | MiMo-V2.5-Pro | 61.4% | estimated ± 2.6 pp, low confidence |
| 51 | MiniMax M3.1 Flash | 61.4% | estimated ± 2.6 pp, low confidence |
| 52 | Qwen3.8-Flash-Next | 61.4% | estimated ± 2.6 pp, low confidence |
| 53 | Beam 1.0 501B-A23B FP8 | 61.4% | estimated ± 2.6 pp, low confidence |
| 54 | Apodex 1.1 | 61.4% | estimated ± 2.6 pp, low confidence |
| 55 | Apodex 1.1 Mini | 61.4% | estimated ± 2.6 pp, low confidence |
| 56 | Claude Opus 5 | 61.4% | estimated ± 2.6 pp, low confidence |
| 57 | Gemini 4 Carbon | 61.4% | estimated ± 2.6 pp, low confidence |
| 58 | Grok 4.5 | 61.4% | estimated ± 2.6 pp, low confidence |
| 59 | Kimi K2.7 Code | 61.4% | estimated ± 2.6 pp, low confidence |
| 60 | Hy3 | 61.4% | estimated ± 2.6 pp, low confidence |
| 61 | Muse Spark 1.2 | 61.4% | estimated ± 2.6 pp, low confidence |
| 62 | Qwen3.7 Max | 61.4% | estimated ± 2.6 pp, low confidence |
| 63 | Claude Opus 4.7 (Adaptive) | 61.4% | estimated ± 2.6 pp, low confidence |
| 64 | Beam 1.0 501B-A23B NVFP4 | 61.4% | estimated ± 2.6 pp, low confidence |
| 65 | GLM-5.2 | 61.3% | estimated ± 2.6 pp, low confidence |
| 66 | Ling 3.0 Flash VL | 61.3% | estimated ± 2.6 pp, low confidence |
| 67 | MiniMax M2.7 | 61.3% | estimated ± 2.6 pp, low confidence |
| 68 | Muse Spark Open | 61.3% | estimated ± 2.6 pp, low confidence |
| 69 | Qwen3.6 Plus | 61.3% | estimated ± 2.6 pp, low confidence |
| 70 | GPT-5 (high) | 61.3% | estimated ± 2.6 pp, low confidence |
| 71 | Beam 1.0 501B-A23B | 61.3% | estimated ± 2.6 pp, low confidence |
| 72 | Claude Opus 4.6 (Adaptive) | 61.3% | estimated ± 2.6 pp, low confidence |
| 73 | Kimi K2.5 (Reasoning) | 61.3% | estimated ± 2.6 pp, low confidence |
| 74 | Muse Spark | 61.3% | estimated ± 2.6 pp, low confidence |
| 75 | Claude Opus 4.8 | 61.3% | estimated ± 2.6 pp, low confidence |
| 76 | Muse Spark 1.1 | 61.3% | estimated ± 2.6 pp, low confidence |
| 77 | Qwen3.5-27B | 61.3% | estimated ± 2.6 pp, low confidence |
| 78 | Claude Opus 4.5 Thinking | 61.3% | estimated ± 2.6 pp, low confidence |
| 79 | Inkling | 61.3% | estimated ± 2.6 pp, low confidence |
| 80 | Qwen3.5 397B (Reasoning) | 61.3% | estimated ± 2.6 pp, low confidence |
| 81 | Qwen3.6-27B | 61.3% | estimated ± 2.6 pp, low confidence |
| 82 | GPT-5.4 mini | 61.3% | estimated ± 2.6 pp, low confidence |
| 83 | Ternary Bonsai 2 27B | 61.3% | estimated ± 2.6 pp, low confidence |
| 84 | GPT-5.4 nano | 61.3% | estimated ± 2.6 pp, low confidence |
| 85 | Grok 4.7 | 61.3% | estimated ± 2.6 pp, low confidence |
| 86 | Quasar 438B | 61.2% | estimated ± 2.6 pp, low confidence |
| 87 | Qwen3.5-122B-A10B | 61.2% | estimated ± 2.6 pp, low confidence |
| 88 | Claude 4.1 Opus Thinking | 61.2% | estimated ± 2.6 pp, low confidence |
| 89 | Gemini 3.5 Flash-Lite | 61.2% | estimated ± 2.6 pp, low confidence |
| 90 | GPT-5 (medium) | 61.2% | estimated ± 2.6 pp, low confidence |
| 91 | Claude Opus 4.7 | 61.2% | estimated ± 2.6 pp, low confidence |
| 92 | GLM-5 | 61.2% | estimated ± 2.6 pp, low confidence |
| 93 | Inkling-Small | 61.2% | estimated ± 2.6 pp, low confidence |
| 94 | MiMo-V2-Omni | 61.2% | estimated ± 2.6 pp, low confidence |
| 95 | o3 | 61.2% | estimated ± 2.6 pp, low confidence |
| 96 | MiMo-V2.6-Flash | 61.2% | estimated ± 2.6 pp, low confidence |
| 97 | Grok 4.1 Fast (Reasoning) | 61.1% | estimated ± 2.6 pp, low confidence |
| 98 | GLM-5.1 | 61.1% | estimated ± 2.6 pp, low confidence |
| 99 | Grok 4 Fast (Reasoning) | 61.1% | estimated ± 2.6 pp, low confidence |
| 100 | Step 3.7 Flash | 61.1% | estimated ± 2.6 pp, low confidence |
| 101 | Ling 3.0 Flash | 61.1% | estimated ± 2.6 pp, low confidence |
| 102 | Qwen3.7 Plus | 61.1% | estimated ± 2.6 pp, low confidence |
| 103 | Qwen3.5-35B-A3B | 61.0% | estimated ± 2.6 pp, low confidence |
| 104 | GLM-5-Turbo | 61.0% | estimated ± 2.6 pp, low confidence |
| 105 | Qwen3.6-35B-A3B | 61.0% | estimated ± 2.6 pp, low confidence |
| 106 | GLM-4.7 | 61.0% | estimated ± 2.6 pp, low confidence |
| 107 | Solar Pro 4 | 61.0% | estimated ± 2.6 pp, low confidence |
| 108 | Claude Opus 4.5 | 61.0% | estimated ± 2.6 pp, low confidence |
| 109 | Ling 3.0 Flash Sante | 61.0% | estimated ± 2.6 pp, low confidence |
| 110 | GLM-5V-Turbo | 60.9% | estimated ± 2.6 pp, low confidence |
| 111 | Gemma 4 31B | 60.9% | estimated ± 2.6 pp, low confidence |
| 112 | Nemotron 3.5 Bolt 74B-A9B | 60.9% | estimated ± 2.6 pp, low confidence |
| 113 | Claude 4 Opus Thinking | 60.9% | estimated ± 2.6 pp, low confidence |
| 114 | Gemini 3.5 Flash | 60.9% | estimated ± 2.6 pp, low confidence |
| 115 | GPT-5.1-Codex | 60.9% | estimated ± 2.6 pp, low confidence |
| 116 | GPT-5.1-Codex-Max | 60.9% | estimated ± 2.6 pp, low confidence |
| 117 | Mistral Medium 3.5 128B | 60.9% | estimated ± 2.6 pp, low confidence |
| 118 | Gemini 2.5 Pro | 60.9% | estimated ± 2.6 pp, low confidence |
| 119 | Claude Sonnet 4.6 | 60.8% | estimated ± 2.6 pp, low confidence |
| 120 | GPT-4.1 | 60.8% | estimated ± 2.6 pp, low confidence |
| 121 | MiMo-V2-Pro | 60.8% | estimated ± 2.6 pp, low confidence |
| 122 | Grok 4 | 60.8% | estimated ± 2.6 pp, low confidence |
| 123 | Mercury 2.5 | 60.8% | estimated ± 2.6 pp, low confidence |
| 124 | Nemotron 3.5 Bolt 74B-A9B BF16 | 60.8% | estimated ± 2.6 pp, low confidence |
| 125 | Spark-X2.6-4B | 60.8% | estimated ± 2.6 pp, low confidence |
| 126 | Claude Opus 4.6 | 60.8% | estimated ± 2.6 pp, low confidence |
| 127 | Nemotron 3 Ultra | 60.8% | estimated ± 2.6 pp, low confidence |
| 128 | Hy3 Preview | 60.7% | estimated ± 2.6 pp, low confidence |
| 129 | A.X K2 | 60.7% | estimated ± 2.6 pp, low confidence |
| 130 | Gemma 4 26B A4B | 60.7% | estimated ± 2.6 pp, low confidence |
| 131 | o1 | 60.6% | estimated ± 2.6 pp, low confidence |
| 132 | Grok 4.3 | 60.6% | estimated ± 2.6 pp, low confidence |
| 133 | Gemma 4 12B | 60.6% | estimated ± 2.6 pp, low confidence |
| 134 | Qwen3.5-4B | 60.5% | estimated ± 2.6 pp, low confidence |
| 135 | Gemini 3 Pro | 60.5% | measured |
| 136 | Solar Open 2 | 60.5% | estimated ± 2.6 pp, low confidence |
| 137 | K-Exaone | 60.4% | estimated ± 2.6 pp, low confidence |
| 138 | o4-mini | 60.4% | estimated ± 2.6 pp, low confidence |
| 139 | Ling 3.0 Tiny | 60.3% | estimated ± 2.6 pp, low confidence |
| 140 | MiniCPM5-2B | 60.2% | estimated ± 2.6 pp, low confidence |
| 141 | DeepSeek V3.1 (Reasoning) | 60.0% | estimated ± 2.6 pp, low confidence |
| 142 | K-EXAONE 2.0 | 60.0% | estimated ± 2.6 pp, low confidence |
| 143 | DeepSeek-R1 | 60.0% | estimated ± 2.6 pp, low confidence |
| 144 | Gemini 3 Flash | 59.9% | estimated ± 2.6 pp, low confidence |
| 145 | Grok Code Fast 1 | 59.7% | estimated ± 2.6 pp, low confidence |
| 146 | Kimi K2 | 59.7% | estimated ± 2.6 pp, low confidence |
| 147 | LFM3-1.7B | 59.7% | estimated ± 2.6 pp, low confidence |
| 148 | Command A+ | 59.7% | estimated ± 2.6 pp, low confidence |
| 149 | GPT-OSS 120B | 59.6% | estimated ± 2.6 pp, low confidence |
| 150 | Claude 3.7 Sonnet | 59.6% | estimated ± 2.6 pp, low confidence |
| 151 | Llama 4 Maverick | 59.4% | estimated ± 2.6 pp, low confidence |
| 152 | Qwen3 Max | 59.4% | estimated ± 2.6 pp, low confidence |
| 153 | Gemini 2.5 Flash | 59.4% | estimated ± 2.6 pp, low confidence |
| 154 | Mistral Small 4 | 59.4% | estimated ± 2.6 pp, low confidence |
| 155 | GPT-4o | 59.4% | estimated ± 2.6 pp, low confidence |
| 156 | Nemotron 3.5 Lightning 30B A3B NVFP4 | 59.3% | estimated ± 2.6 pp, low confidence |
| 157 | Granite 4.2 30B | 59.3% | estimated ± 2.6 pp, low confidence |
| 158 | DeepSeek V3.1 | 59.1% | estimated ± 2.6 pp, low confidence |
| 159 | GLM-4.5-Air | 59.1% | estimated ± 2.6 pp, low confidence |
| 160 | DeepSeek V3.2 | 59.0% | estimated ± 2.6 pp, low confidence |
| 161 | Granite 4.2 8B | 58.9% | estimated ± 2.6 pp, low confidence |
| 162 | Claude 4 Sonnet | 58.7% | estimated ± 2.6 pp, low confidence |
| 163 | GPT-4.1 mini | 58.7% | estimated ± 2.6 pp, low confidence |
| 164 | Kimi K2.5 | 58.5% | measured |
| 165 | DeepSeek V3 0324 | 58.3% | estimated ± 2.6 pp, low confidence |
| 166 | Celeris-1 | 57.9% | estimated ± 2.6 pp, low confidence |
| 167 | Trinity-Large-Preview | 57.9% | estimated ± 2.6 pp, low confidence |
| 168 | Trinity-Large-Thinking | 57.9% | estimated ± 2.6 pp, low confidence |
| 169 | North Mini Code | 57.7% | estimated ± 2.6 pp, low confidence |
| 170 | MiMo-V2-Flash | 57.6% | estimated ± 2.6 pp, low confidence |
| 171 | Mistral Large 3 | 57.5% | estimated ± 2.6 pp, low confidence |
| 172 | Nemotron 3 Nano Omni 30B A3B | 57.5% | measured |
| 173 | GPT-OSS 20B | 57.3% | estimated ± 2.6 pp, low confidence |
| 174 | Solar Pro 3 | 56.8% | estimated ± 2.6 pp, low confidence |
| 175 | Gemini 2.5 Flash-Lite | 56.7% | estimated ± 2.6 pp, low confidence |
| 176 | Gemma 4 E4B | 56.7% | estimated ± 2.6 pp, low confidence |
| 177 | Gemini 2.0 Flash | 56.5% | estimated ± 2.6 pp, low confidence |
| 178 | Grok 4.1 Fast | 56.5% | estimated ± 2.6 pp, low confidence |
| 179 | Ling 2.6 Flash | 56.5% | estimated ± 2.6 pp, low confidence |
| 180 | Mistral Medium 3 | 56.5% | estimated ± 2.6 pp, low confidence |
| 181 | DeepSeek V3 | 56.0% | estimated ± 2.6 pp, low confidence |
| 182 | Mistral Small 4 (Reasoning) | 55.8% | estimated ± 2.6 pp, low confidence |
| 183 | Claude 3 Haiku | 55.6% | estimated ± 2.6 pp, low confidence |
| 184 | Llama 4 Scout | 55.6% | estimated ± 2.6 pp, low confidence |
| 185 | Claude 3.5 Haiku | 55.5% | estimated ± 2.6 pp, low confidence |
| 186 | QwQ-32B | 55.3% | estimated ± 2.6 pp, low confidence |
| 187 | GLM-4.6 | 55.2% | estimated ± 2.6 pp, low confidence |
| 188 | Llama 3.1 405B | 54.8% | estimated ± 2.6 pp, low confidence |
| 189 | Granite 4.2 3B | 54.5% | estimated ± 2.6 pp, low confidence |
| 190 | Mistral Small 3.1 | 53.7% | estimated ± 2.6 pp, low confidence |
| 191 | Nova Pro | 53.1% | estimated ± 2.6 pp, low confidence |
| 192 | GPT-4.1 nano | 52.8% | estimated ± 2.6 pp, low confidence |
| 193 | Devstral Small | 52.0% | estimated ± 2.6 pp, low confidence |
| 194 | Gemma 4 E2B | 50.5% | estimated ± 2.6 pp, low confidence |
| 195 | Llama 3.3 70B | 50.0% | estimated ± 2.6 pp, low confidence |
| 196 | Ultravox v0.6 Llama 3.3 70B | 50.0% | estimated ± 2.6 pp, low confidence |
| 197 | Nemotron 3 Nano 30B | 45.2% | estimated ± 2.6 pp, low confidence |
| 198 | DeepSeek R1 Distill Qwen 32B | 42.2% | estimated ± 2.6 pp, low confidence |
| 199 | Gemma 3 27B | 39.5% | estimated ± 2.6 pp, low confidence |
| 200 | Granite-4.0-H-1B | 38.9% | estimated ± 2.6 pp, low confidence |
| 201 | LFM2.5-2.6B | 35.6% | estimated ± 2.6 pp, low confidence |
| 202 | Llama 3.1 70B | 32.5% | estimated ± 2.6 pp, low confidence |
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 Hard
- Terminal-Bench 2.1
- Terminal-Bench 2.1 (Vals AI)
Agentic · tools
- AA Agentic Index
- AA-AnalystAgent
- AA AutomationBench
- CWE-Bench-AA
- AA Cyber Index
- DeepsecBench-AA
- 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
- Finance Agent v2
- GDPval-AA
- Gert Labs
- Hermes Bench
- Hermes Index
- HLE w/ tools
- JobBench
- MazeBench
- MCP Atlas
- MCP-Tasks
- OSWorld 2.1
- OSWorld 2.1 Offline
- OSWorld-Verified
- OSWorld 2.0
- PinchBench
- QwenClawBench
- ResearchClawBench
- SkillsBench
- τ²-bench results
- τ³-bench results
- Toolathlon
- Toolathlon-Verified
- VITA-Bench
- WideResearch
Coding
- AA Coding Index
- AA LiveCodeBench
- AA-SciCode
- Bug Hunt Bench
- CursorBench 3.1
- CursorBench 3.2
- CursorBench 4.0
- DeepSWE
- FrontierCode 1.1 Main
- FrontierCode 2.0
- FrontierCode 1.1 Extended
- FrontierSWE v2
- LiveCodeBench
- LiveCodeBench Pro
- LiveCodeBench v6
- NL2Repo
- OpenHarmony Bench
- PostTrainBench v1.1
- ProgramBench
- React Native Evals
- SciCode
- SWE-Atlas Codebase QnA
- SWE-bench Pro
- SWE-bench Verified
- SWE-Rebench
- SWE-Marathon
- Vals LiveCodeBench
- Vals SWE-bench
- Vibe Code Bench
- VulcanBench v3
Math
- AIME 2024
- AIME 2025
- AIME26
- 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
- AA MMLU-Pro
- ARC-AGI-2
- ARC-AGI-3 Provider Adapter
- ARC-AGI-1
- ARC-AGI-3
- Artificial Analysis Intelligence Index
- BBH
- BioMysteryBench (human-difficult)
- BioMysteryBench (human-solvable)
- C-Eval
- CritPt
- GPQA
- GPQA Diamond
- GraphWalks BFS 256K–1M
- 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
- STEMBench
- SuperGPQA
- Vals GPQA Diamond
- Vals MMLU-Pro
Instruction following
Multilingual
Vision & documents
- AA-MMMU-Pro
- AI2D_TEST
- BabyVision
- CC-OCR
- Chartography (tools)
- CharXiv
- CharXiv w/o tools
- CountBench
- DynaMath
- ERQA
- LVBench
- MathVision
- MathVision w/ Python
- MedXpertQA (MM)
- MLVU (M-Avg)
- MMMU
- MMMU-Pro
- MMMU-Pro w/ Python
- MMVU
- OCRBench V2
- OfficeQA Pro
- OmniDocBench 1.5
- ParseBench
- RealWorldQA
- RefCOCO (avg)
- ScreenSpot Pro
- SimpleVQA
- SWE-bench Multimodal
- Video-MME (w/o subtitle)
- Video-MME (with subtitle)
- VideoMMMU
- V*
- We-Math
- ZeroBench