MMMU Pro
University-level questions that require reading charts, diagrams, photographs and documents alongside text.
Multimodal reasoningMMMU team93 models run by Vals AIUpdated 1 September 2026
How it is run
Questions span six disciplines and thirty subjects, with images as the primary evidence rather than decoration.
What it cannot tell you
The board is crowded at the top: the top ten runs sit within three points of each other. Text-heavy models can score well here without strong image generation or editing ability.
- Leader
- Claude Fable 5.190.64% · Vals AI's own list
- Top-ten spread
- 2.43%Leader minus tenth place, or last place in smaller fields.
- Median
- 87.31%of the 20 models on this site
- No published result
- 26of 46 models on this site
Independent leaderboard · Vals AI
Top 93 of 93 runs by Vals AI, one harness.
| # | Model | Score |
|---|---|---|
| 1 | Claude Fable 5.1 | 90.64% |
| 2 | Claude Opus 5 | 89.88% |
| 3 | Claude Fable 5 | 89.31% |
| 4 | Gemini 3.8 Flash | 89.08% |
| 5 | Gemini 3.7 Flash | 88.96% |
| 6 | GPT-5.6 Sol | 88.84% |
| 7 | Gemini 3.6 Flash | 88.38% |
| 8 | Gemini 3.5 Flash | 88.27% |
| 8 | GPT-5.5 | 88.27% |
| 10 | Gemini 3.1 Pro | 88.21% |
| 11 | Kimi K3 | 88.15% |
| 12 | Qwen3.8 Max | 88.04% |
| 13 | Gemini 3 Flash Preview | 87.63% |
| 14 | Gemini 3 Pro Preview | 87.51% |
| 14 | Gpt 5.4 2026 03 05 | 87.51% |
| 16 | Muse Spark | 87.40% |
| 17 | Gpt 5.2 2025 12 11 | 86.67% |
| 18 | Muse Spark 1 1 | 86.59% |
| 18 | Claude Opus 4.8 | 86.59% |
| 20 | GPT-5.6 Terra | 86.47% |
| 21 | Kimi K2.6 | 86.30% |
| 22 | Muse Spark 1.2 | 86.13% |
| 23 | GLM 5.3 Flash | 86.01% |
| 24 | Claude Opus 4.7 | 85.55% |
| 25 | GPT-5.6 Luna | 85.03% |
| 26 | Kimi K2.5 Thinking | 84.34% |
| 27 | Qwen3.6 Plus | 84.16% |
| 28 | Qwen3.8 27b | 83.93% |
| 29 | Claude Opus 4 6 Thinking | 83.87% |
| 30 | Gemini 3.5 Flash Lite | 83.58% |
| 30 | Claude Sonnet 4 6 | 83.58% |
| 32 | Grok 4.20 0309 Reasoning | 83.47% |
| 33 | Gpt 5.1 2025 11 13 | 83.18% |
| 34 | Grok 4.3 | 83.06% |
| 35 | Claude Sonnet 5 | 83.01% |
| 36 | Claude Opus 4 5 20251101 Thinking | 82.95% |
| 37 | Gemini 3.1 Flash Lite Preview | 82.49% |
| 38 | Qwen3.5 Flash | 81.91% |
| 39 | Gpt 5 2025 08 07 | 81.50% |
| 40 | Gemini 2.5 Pro Exp 03 25 | 81.34% |
| 41 | MiniMax M3 | 81.16% |
| 42 | Claude Opus 4 5 20251101 | 81.10% |
| 43 | Gemini 2.5 Flash Preview 09 2025 Thinking | 80.75% |
| 44 | O3 2025 04 16 | 80.42% |
| 45 | Mimo V2.5 | 80.00% |
| 46 | O4 Mini 2025 04 16 | 79.67% |
| 47 | Gemini 2.5 Flash Preview 09 2025 | 79.48% |
| 48 | Claude Sonnet 4 5 20250929 Thinking | 79.31% |
| 49 | Gpt 5.4 Mini 2026 03 17 | 79.25% |
| 50 | Gpt 5 Mini 2025 08 07 | 78.91% |
| 51 | Inkling | 78.50% |
| 52 | Claude Opus 4 1 20250805 Thinking | 77.51% |
| 53 | O1 2024 12 17 | 77.41% |
| 54 | Grok 4 0709 | 76.27% |
| 55 | Gemini 2.5 Flash Lite Preview 09 2025 Thinking | 75.43% |
| 56 | Claude 3 7 Sonnet 20250219 Thinking | 75.10% |
| 57 | Claude Sonnet 4 20250514 Thinking | 74.93% |
| 58 | Claude Opus 4 1 20250805 | 73.72% |
| 59 | Gpt 5.4 Nano 2026 03 17 | 73.58% |
| 60 | Claude Opus 4 20250514 | 73.31% |
| 61 | Grok 4 Fast Reasoning | 72.78% |
| 62 | Grok 4 1 Fast Reasoning | 72.66% |
| 63 | Gemini 2.5 Flash Lite Preview 09 2025 | 72.54% |
| 64 | Claude Sonnet 4 20250514 | 72.39% |
| 64 | Gpt 4.1 2025 04 14 | 72.39% |
| 66 | Gemini 2.5 Flash Preview 04 17 Thinking | 71.92% |
| 67 | Llama4 Maverick Instruct Basic | 71.69% |
| 68 | Claude 3 7 Sonnet 20250219 | 71.52% |
| 69 | Gpt 5 Nano 2025 08 07 | 70.94% |
| 70 | Gpt 4.1 Mini 2025 04 14 | 70.54% |
| 71 | Gemini 2.0 Flash 001 | 69.79% |
| 72 | Claude 3 5 Sonnet 20241022 | 68.80% |
| 73 | Gemini 2.5 Flash Preview 04 17 | 68.00% |
| 74 | Mistral Large 2512 | 66.19% |
| 75 | Gemini 1.5 Pro 002 | 65.51% |
| 76 | Magistral Small 2509 | 65.20% |
| 77 | Magistral Medium 2509 | 64.57% |
| 78 | Gpt 4o 2024 08 06 | 64.01% |
| 79 | Grok 4 1 Fast Non Reasoning | 63.70% |
| 80 | Grok 4 Fast Non Reasoning | 63.41% |
| 81 | Mistral Medium 2505 | 62.97% |
| 82 | Gpt 4o 2024 11 20 | 62.16% |
| 83 | Grok 4.5 | 61.79% |
| 84 | Mistral Small 2503 | 60.08% |
| 85 | Llama 4 Scout 17B 16E Instruct | 58.75% |
| 86 | Grok 2 Vision 1212 | 57.25% |
| 87 | Gemini 1.5 Flash 002 | 57.19% |
| 88 | Gpt 4o Mini 2024 07 18 | 56.56% |
| 89 | Gpt 4.1 Nano 2025 04 14 | 55.06% |
| 90 | Llama 3.2 90B Vision Instruct Turbo | 48.07% |
| 91 | Claude Haiku 4 5 20251001 Thinking | 46.07% |
| 92 | Llama 3.2 11B Vision Instruct Turbo | 38.82% |
| 93 | Qwen3.5 Plus Thinking | 22.78% |
Artificial Analysis results by release
Highest published effort per model release, grouped here from AA’s public results. Grouping ranks are derived here; source settings and model-publisher reports stay separate.
| # | Model | Score | Effort |
|---|---|---|---|
| 1 | Claude Opus 5.5 | 87.69% | max |
| 2 | GPT-6 Astra | 86.88% | max |
| 3 | GPT-6.1 Sol | 85.95% | max |
| 4 | Gemini 3.8 Flash | 85.61% | high |
| 5 | Gemini 3.7 Flash | 85.49% | high |
| 6 | Claude Opus 5 | 84.74% | max |
| 7 | Gemini 3.5 Flash | 84.28% | high |
| 8 | GPT-5.6 Sol | 83.41% | max |
| 9 | GPT-6 Sol | 83.29% | max |
| 10 | Gemini 3.6 Flash | 83.24% | high |
| 11 | Qwen3.8 Max (0902) | 82.77% | Not specified |
| 12 | Gemini 3.1 Pro Preview | 82.43% | Not specified |
| 13 | Qwen3.8 Max | 82.31% | Not specified |
| 14 | Muse Spark 1.3 | 82.02% | xhigh |
| 15 | GPT-5.6 Terra | 80.69% | max |
| 16 | Kimi K3 | 80.52% | max |
| 16 | Muse Spark | 80.52% | Not specified |
| 18 | Qwen3.7 Plus | 80.46% | Not specified |
| 19 | Grok 4.5 | 80.40% | high |
| 20 | Gemini 3 Pro Preview | 80.17% | high |
| 21 | GPT-5.5 | 79.88% | xhigh |
| 22 | Qwen3.8-Flash-Next | 79.77% | Not specified |
| 23 | Kimi K2.6 | 79.36% | Not specified |
| 24 | Apodex 1.1 | 79.19% | Not specified |
| 25 | Gemini 3.5 Flash-Lite | 79.02% | high |
| 26 | Ling-3.0-flash-VL | 78.96% | Not specified |
| 27 | Claude Opus 4.7 | 78.84% | max |
| 28 | GPT-5.6 Luna | 78.55% | max |
| 28 | MiniMax-M3 | 78.55% | Not specified |
| 28 | Gemini 3 Flash Preview | 78.55% | Not specified |
| 31 | GPT-5.3 Codex | 78.50% | xhigh |
| 32 | GPT-5.4 | 78.44% | xhigh |
| 33 | Grok 4.3 | 78.09% | high |
| 34 | Qwen3.6 Plus | 77.98% | Not specified |
| 35 | Qwen3.5 397B A17B | 77.28% | Not specified |
| 35 | Claude Sonnet 5 | 77.28% | max |
| 37 | DeepSeek V4.1 Flash | 76.99% | max |
| 38 | Grok Build 0.1 0616 | 76.53% | Not specified |
| 39 | Step 5 Preview | 76.36% | Not specified |
| 40 | GPT-5.2 Codex | 76.30% | xhigh |
| 40 | Qwen3.8 27B | 76.30% | xhigh |
| 42 | GPT-6 Luna | 75.55% | max |
| 42 | Gemini 3.1 Flash-Lite | 75.55% | high |
| 44 | GPT-5.1 | 75.49% | high |
| 45 | MiMo-V2.5 | 75.43% | Not specified |
| 45 | Claude Opus 4.6 | 75.43% | max |
| 47 | Kimi K2.5 | 75.38% | Not specified |
| 48 | Step 3.7 Flash | 75.32% | high |
| 49 | Qwen3.5 27B | 75.03% | Not specified |
| 50 | Gemini 2.5 Pro | 74.91% | Not specified |
| 51 | DeepSeek V4 Flash Vision | 74.80% | max |
| 52 | GPT-5.2 | 74.57% | medium |
| 53 | Muse Glimmer | 74.34% | high |
| 54 | GPT-5 | 74.22% | high |
| 55 | Claude Opus 4.5 | 74.05% | Not specified |
| 55 | Inkling Small | 74.05% | Not specified |
| 57 | MiMo-V2-Omni-0327 | 73.93% | Not specified |
| 58 | GPT-5 Codex | 73.82% | high |
| 59 | Inkling | 73.47% | xhigh |
| 60 | Gemma 4 31B | 73.41% | Not specified |
| 61 | Claude Sonnet 4.6 | 73.29% | max |
| 61 | GPT-5.4 mini | 73.29% | xhigh |
| 63 | Grok 4.20 0309 | 73.18% | Not specified |
| 64 | MiMo-V2.6-Flash | 73.06% | Not specified |
| 65 | GLM 5V Turbo | 72.77% | Not specified |
| 66 | Qwen3.5 35B A3B | 72.66% | Not specified |
| 67 | GPT-5.1 Codex | 72.49% | high |
| 68 | Qwen3.6 27B | 71.73% | Not specified |
| 69 | Qwen3.6 35B A3B | 71.04% | Not specified |
| 70 | Qwen3.5 Omni Plus | 70.52% | Not specified |
| 71 | Qwen3.5 122B A10B | 70.29% | Not specified |
| 72 | Gemini 2.5 Flash Preview (Sep '25) | 70.23% | Not specified |
| 73 | o3 | 70.06% | Not specified |
| 73 | GPT-5 mini | 70.06% | high |
| 75 | MiMo-V2-Omni | 69.88% | Not specified |
| 76 | Gemma 4 26B A4B | 69.25% | Not specified |
| 76 | o4-mini | 69.25% | high |
| 78 | Gemini 2.5 Flash | 69.08% | Not specified |
| 79 | GPT-5.1 Codex mini | 69.02% | high |
| 80 | Grok 4 | 68.84% | Not specified |
| 81 | Claude 4.5 Sonnet | 68.73% | Not specified |
| 81 | Qwen3 VL 235B A22B (Reasoning) | 68.73% | Not specified |
| 83 | Doubao Seed Code | 68.09% | Not specified |
| 84 | Claude 4.1 Opus | 67.92% | Not specified |
| 85 | Qwen3 VL 235B A22B Instruct | 67.57% | Not specified |
| 86 | EXAONE 4.5 33B | 67.34% | Not specified |
| 87 | Qwen3.5 9B | 66.76% | Not specified |
| 88 | DiffusionGemma 26B A4B | 66.53% | Not specified |
| 89 | GPT-5.4 nano | 65.38% | xhigh |
| 90 | Grok 4.20 0309 v2 | 64.91% | Not specified |
| 91 | Mistral Medium 3.5 | 64.86% | high |
| 92 | Qwen3.5 Omni Flash | 64.74% | Not specified |
| 93 | ERNIE 5.0 Thinking Preview | 64.57% | Not specified |
| 94 | Nova 2.0 Pro Preview | 64.51% | medium |
| 95 | Qwen3 VL 32B Instruct | 64.28% | Not specified |
| 96 | JT-4.1 Flash 236B A21B | 64.10% | Not specified |
| 97 | Step3 VL 10B | 63.99% | Not specified |
| 98 | Nova 2.0 Lite | 63.76% | high |
| 99 | Gemini 2.5 Flash-Lite Preview (Sep '25) | 63.41% | Not specified |
| 99 | Qwen3 VL 32B (Reasoning) | 63.41% | Not specified |
| 101 | Grok 4.1 Fast | 63.29% | Not specified |
| 102 | Command A+ | 63.24% | Not specified |
| 103 | Qwen3 VL 30B A3B Instruct | 62.14% | Not specified |
| 103 | Llama 4 Maverick | 62.14% | Not specified |
| 105 | Qwen3.5 4B | 62.08% | Not specified |
| 106 | Gemma 4 12B | 61.97% | Not specified |
| 106 | Gemini 2.5 Flash Preview | 61.97% | Not specified |
| 108 | Nova 2.0 Omni | 61.91% | medium |
| 109 | Qwen3 VL 30B A3B (Reasoning) | 61.85% | Not specified |
| 110 | Claude 4 Sonnet | 61.79% | Not specified |
| 111 | GPT-4.1 | 61.21% | Not specified |
| 112 | GPT-5 nano | 60.98% | high |
| 113 | Qwen3 Omni 30B A3B (Reasoning) | 60.23% | Not specified |
| 114 | Claude 3.7 Sonnet | 60.06% | Not specified |
| 115 | Magistral Medium 1.2 | 59.65% | Not specified |
| 116 | GPT-4.1 mini | 58.73% | Not specified |
| 117 | Claude 4.5 Haiku | 58.55% | Not specified |
| 118 | Apriel-v1.5-15B-Thinker | 57.05% | Not specified |
| 119 | Qwen3 VL 8B (Reasoning) | 56.65% | Not specified |
| 120 | GPT-4o (Aug '24) | 56.30% | Not specified |
| 121 | Mistral Large 3 | 55.66% | Not specified |
| 122 | Qwen3 Omni 30B A3B Instruct | 55.49% | Not specified |
| 122 | Magistral Small 1.2 | 55.49% | Not specified |
| 124 | Gemini 1.5 Pro (Sep '24) | 55.03% | Not specified |
| 125 | Mistral Medium 3.1 | 54.16% | Not specified |
| 126 | Gemini 2.5 Flash-Lite | 53.99% | Not specified |
| 127 | Nemotron 3 Nano Omni 30B A3B Reasoning | 53.18% | Not specified |
| 128 | Mistral Medium 3 | 53.01% | Not specified |
| 129 | Llama 4 Scout | 52.95% | Not specified |
| 130 | NVIDIA Nemotron Nano 12B v2 VL | 52.89% | Not specified |
| 131 | Qwen3 VL 4B (Reasoning) | 52.02% | Not specified |
| 132 | Gemma 4 E4B | 51.21% | Not specified |
| 133 | Pixtral Large | 50.58% | Not specified |
| 134 | Ministral 3 14B | 49.83% | Not specified |
| 135 | Gemini 1.5 Flash (Sep '24) | 48.38% | Not specified |
| 136 | Grok 4 Fast | 48.09% | Not specified |
| 137 | Gemma 3 27B Instruct | 48.03% | Not specified |
| 138 | Mistral Small 3.2 | 47.98% | Not specified |
| 139 | Qwen3 VL 8B Instruct | 47.34% | Not specified |
| 140 | Mistral Small 4 | 46.47% | Not specified |
| 141 | Ministral 3 8B | 45.95% | Not specified |
| 142 | Claude 3.5 Haiku | 45.61% | Not specified |
| 143 | Devstral Small 2 | 44.62% | Not specified |
| 144 | Nova Pro | 44.34% | Not specified |
| 145 | Qwen3 VL 4B Instruct | 43.87% | Not specified |
| 146 | GLM-4.5V | 42.77% | Not specified |
| 147 | Qwen3.5 2B | 42.66% | Not specified |
| 148 | GLM-4.6V | 42.20% | Not specified |
| 149 | Gemma 4 E2B | 41.79% | Not specified |
| 150 | GPT-4o mini | 41.50% | Not specified |
| 151 | GPT-4.1 nano | 40.12% | Not specified |
| 152 | Llama 3.2 Instruct 90B (Vision) | 39.48% | Not specified |
| 153 | Ministral 3 3B | 38.09% | Not specified |
| 154 | MiniCPM-V 4.6 1.3B | 37.92% | Not specified |
| 155 | Nova Lite | 37.80% | Not specified |
| 156 | Gemma 3 12B Instruct | 37.51% | Not specified |
| 157 | Molmo2-8B | 37.46% | Not specified |
| 158 | Gemini 1.5 Flash-8B | 36.53% | Not specified |
| 159 | Claude 3 Haiku | 30.81% | Not specified |
| 160 | Gemma 3 4B Instruct | 29.94% | Not specified |
| 161 | Llama 3.2 Instruct 11B (Vision) | 29.31% | Not specified |
| 162 | LFM2.5-VL-1.6B | 26.53% | Not specified |
| 163 | Gemma 3n E4B Instruct | 26.24% | Not specified |
| 164 | Qwen3.5 0.8B | 25.66% | Not specified |
| 165 | Molmo 7B-D | 24.51% | Not specified |
| 166 | Phi-4 Multimodal Instruct | 14.51% | Not specified |
No published result tracked
26 of the 46 models on this site have no published number on this benchmark, so their cells in the home table stay blank rather than reading as zero.
Step 5 PreviewGPT-6 LunaQwen3.8 2.4T A95BQwen3.8-Flash-NextMiMo-V2.6-FlashQwen3.8 Max (0902)Grok 4.6GPT-5.4DeepSeek V4.1 Flash 1017GLM-LegalClaude Haiku 5.5BeamGemini 4 ArgonClaude Sonnet 5.5Claude Opus 5.5GPT-6 AstraGPT-6 SolGPT-6.1 SolGrok 4.7DeepSeek V4.1 FlashDeepSeek V4 ProGLM 5.3Muse Spark 1.3 MaxHunyuan Hy4 PreviewMiMo V2.6 ProMistral Medium 3.5
Sources
Every page used for this benchmark, read on 17 October 2026.
- Benchmark site or the page that publishes itmmmu-benchmark.github.io/
- Independent resultswww.vals.ai/benchmarks/mmmu