MMMU Pro (Vals AI) leaderboard
- Claude Fable 5.1 leads
- 90.6%
- models measured
- 63
- lead over #2
- +0.8 pts
- median
- 83.1%
Leader on MMMU Pro (Vals AI) · read October 22
On October 22, 2026, 66 of our catalog's models had a MMMU Pro (Vals AI) result, and Anthropic Claude Fable 5.1 led with 90.6%, ahead of Anthropic Claude Opus 5 at 89.9%.
Questions that need an image and its text read together, across many college subjects. Share answered correctly, as run by Vals AI.
Role in the index
MMMU Pro (Vals AI) is shown as published and does not feed any of our indexes. Read the methodology
| # | Model | Result | Latency, as printed | Evidence | Coding index | Output $/Mtok |
|---|---|---|---|---|---|---|
| 1 | Claude Fable 5.1Anthropic | 90.6% | 37 s | 84.2 | $50.00 | |
| 2 | Claude Opus 5Anthropic | 89.9% | 30 s | 83.0 | $25.00 | |
| 3 | Claude Fable 5Anthropic | 89.3% | 1m 01s | 79.3 | $50.00 | |
| 4 | Gemini 3.8 FlashGoogle | 89.1% | 12 s | 61.6 | $3.75 | |
| 5 | Gemini 3.7 FlashGoogle | 89.0% | 6.3 s | 58.0 | $3.75 | |
| 6 | GPT-5.6 SolOpenAI | 88.8% | 26 s | 73.5 | $20.00 | |
| 7 | Gemini 3.6 FlashGoogle | 88.4% | 11 s | 62.5 | $3.75 | |
| 8 | Gemini 3.5 FlashGoogle | 88.3% | 12 s | 39.8 | $9.00 | |
| 9 | GPT-5.5OpenAI | 88.3% | 54 s | 71.3 | $30.00 | |
| 10 | Gemini 3.1 Pro PreviewGoogle | 88.2% | 1m 17s | 32.3 | $12.00 | |
| 11 | Kimi K3Moonshot AI | 88.2% | 1m 47s | 62.3 | $15.00 | |
| 12 | Qwen3.8 Max (0902)Qwen | 88.0% | 44 s | 62.4 | $6.00 | |
| 13 | Gemini 3 Flash PreviewGoogle | 87.6% | 28 s | 43.7 | $3.00 | |
| 14 | GPT-5.4OpenAI | 87.5% | 1m 06s | 64.4 | $15.00 | |
| 15 | GPT-5.2OpenAI | 86.7% | 1m 36s | 55.9 | $14.00 | |
| 16 | Claude Opus 4.8Anthropic | 86.6% | 49 s | 65.0 | $25.00 | |
| 17 | Muse Spark 1.1Meta | 86.6% | 32 s | 70.3 | $4.25 | |
| 18 | GPT-5.6 TerraOpenAI | 86.5% | 13 s | 63.9 | $12.00 | |
| 19 | Kimi K2.6Moonshot AI | 86.3% | 4m 02s | 52.1 | $4.00 | |
| 20 | Muse Spark 1.2Meta | 86.1% | 1m 14s | 55.7 | $4.25 | |
| 21 | GLM 5.3 FlashZ.ai | 86.0% | 28 s | 57.7 | $0.50 | |
| 22 | Claude Opus 4.7Anthropic | 85.5% | 1m 11s | 67.5 | $25.00 | |
| 23 | GPT-5.6 LunaOpenAI | 85.0% | 35 s | 58.0 | $1.20 | |
| 24 | Kimi K2.5Moonshot AI | 84.3% | 1m 50s | 38.6 | $2.85 | |
| 25 | Qwen3.6 PlusQwen | 84.2% | 1m 52s | 44.1 | $1.95 | |
| 26 | Qwen3.8 27BQwen | 83.9% | 43 s | 74.2 | $2.55 | |
| 27 | Claude Opus 4.6Anthropic | 83.9% | 1m 59s | 57.2 | $25.00 | |
| 28 | Claude Sonnet 4.6Anthropic | 83.6% | 2m 12s | 57.3 | $15.00 | |
| 29 | Gemini 3.5 Flash LiteGoogle | 83.6% | 7.8 s | 49.0 | $2.50 | |
| 30 | Grok 4.20xAI | 83.5% | 44 s | 33.3 | $2.50 | |
| 31 | GPT-5.1OpenAI | 83.2% | 1m 15s | 40.4 | $10.00 | |
| 32 | Grok 4.3xAI | 83.1% | 2m 31s | 37.4 | $2.50 | |
| 33 | Claude Sonnet 5Anthropic | 83.0% | 19 s | 57.7 | $10.00 | |
| 34 | Claude Opus 4.5Anthropic | 83.0% | 1m 15s | 44.0 | $25.00 | |
| 35 | Gemini 3.1 Flash Lite PreviewGoogle | 82.5% | 12 s | 22.7 | $1.50 | |
| 36 | Qwen3.5-FlashQwen | 81.9% | 46 s | 41.1 | $0.26 | |
| 37 | GPT-5OpenAI | 81.5% | 1m 12s | 38.8 | $10.00 | |
| 38 | MiniMax M3MiniMax | 81.2% | 1m 08s | 39.2 | $1.20 | |
| 39 | o3OpenAI | 80.4% | 39 s | Not benchmarked | $8.00 | |
| 40 | MiMo-V2.5Xiaomi | 80.0% | 42 s | 50.0 | $0.336 | |
| 41 | o4 MiniOpenAI | 79.7% | 20 s | Not benchmarked | $4.40 | |
| 42 | Claude Sonnet 4.5Anthropic | 79.3% | 56 s | 39.5 | $15.00 | |
| 43 | GPT-5.4 MiniOpenAI | 79.3% | 51 s | 35.8 | $4.50 | |
| 44 | GPT-5 MiniOpenAI | 78.9% | 27 s | 34.2 | $2.00 | |
| 45 | T InklingThinking Machines | 78.5% | 2m 04s | 31.5 | $4.05 | |
| 46 | Claude Opus 4.1Anthropic | 77.5% | 2.5 s | Not benchmarked | $75.00 | |
| 47 | o1OpenAI | 77.4% | 26 s | Not benchmarked | $60.00 | |
| 48 | Gemini 2.5 Flash LiteGoogle | 75.4% | 26 s | Not benchmarked | $0.40 | |
| 49 | GPT-5.4 NanoOpenAI | 73.6% | 22 s | 44.8 | $1.25 | |
| 50 | GPT-4.1OpenAI | 72.4% | 12 s | Not benchmarked | $8.00 | |
| 51 | Llama 4 MaverickMeta Llama | 71.7% | 7.9 s | Not benchmarked | $1.00 | |
| 52 | GPT-5 NanoOpenAI | 70.9% | 31 s | Not benchmarked | $0.40 | |
| 53 | GPT-4.1 MiniOpenAI | 70.5% | 8.2 s | Not benchmarked | $1.60 | |
| 54 | Gemini 2.0 FlashRetiredGoogle | 69.8% | 45 s | Not benchmarked | $0.40 | |
| 55 | Mistral Large 3 2512Mistral AI | 66.2% | 14 s | Not benchmarked | $1.50 | |
| 56 | Gemini 1.5 ProRetiredGoogle | 65.5% | 44 s | Not benchmarked | $5.00 | |
| 57 | GPT-4o (2024-08-06)OpenAI | 64.0% | 48 s | Not benchmarked | $10.00 | |
| 58 | Mistral Medium 3.1Mistral AI | 63.0% | 13 s | Not benchmarked | $2.00 | |
| 59 | GPT-4o (2024-11-20)OpenAI | 62.2% | 46 s | Not benchmarked | $10.00 | |
| 60 | Grok 4.5xAI | 61.8% | 11m 55s | 58.6 | $6.00 | |
| 61 | Llama 4 ScoutMeta Llama | 58.8% | 11 s | Not benchmarked | $0.50 | |
| 62 | Gemini 1.5 FlashRetiredGoogle | 57.2% | 41 s | Not benchmarked | $0.30 | |
| 63 | GPT-4o-miniOpenAI | 56.6% | 45 s | Not benchmarked | $0.60 | |
| 64 | GPT-4.1 NanoOpenAI | 55.0% | 7.3 s | Not benchmarked | $0.40 | |
| 65 | Claude Haiku 4.5Anthropic | 46.1% | 40 s | 35.6 | $5.00 | |
| 66 | Qwen3.5 Plus 2026-04-20Qwen | 22.8% | 2m 15s | 39.5 | $1.80 |
Only models in our catalog are listed, and each appears once with its best published run. 137 rows were read from the source on October 22, 2026, and rows that match no catalog model are left out. Every other column is a figure the board publishes for the same run.