MMLU-Pro
Harder, reasoning-heavy successor to MMLU. Run by Vals AI.
- Published by
- Vals AI
- Category
- Knowledge
- Index weight
- 1.0
- Models
- 129
- Data as of
- 9 Sept 2026
Vals AI (vals.ai).
- 1Claude Fable 5.192.4%
- 2Claude Opus 591.6%
- 3Claude Fable 591.5%
- 4Gemini 3.1 Pro91.0%
- 5Gemini 3.8 Flash90.2%
- 6Gemini 3.7 Flash90.1%
- 7Gemini 3 Pro90.1%
- 8Claude Opus 4.789.9%
- 9Claude Opus 4.889.6%
- 10Gemini 3.5 Flash89.5%
- 11Grok 4.689.4%
- 12Qwen3 789.3%
- 13Gemini 3.6 Flash89.3%
- 14Grok 4.589.2%
- 15Claude Opus 4.689.1%
129 of 129
| # | |||||
|---|---|---|---|---|---|
| 1 |  Claude Fable 5.1Anthropic | 92.4% | – | 70.0 | – |
| 2 | Claude Opus 5Anthropic | 91.6% | – | 70.0 | – |
| 3 | Claude Fable 5Anthropic | 91.5% | – | 70.4 | – |
| 4 |  Gemini 3.1 ProGoogle | 91.0% | – | 60.7 | – |
| 5 | Gemini 3.8 FlashGoogle | 90.2% | – | 60.1 | – |
| 6 | Gemini 3.7 FlashGoogle | 90.1% | – | 60.8 | – |
| 7 | Gemini 3 ProGoogle | 90.1% | – | 61.6 | – |
| 8 | Claude Opus 4.7Anthropic | 89.9% | – | 65.5 | – |
| 9 | Claude Opus 4.8Anthropic | 89.6% | – | 64.7 | – |
| 10 | Gemini 3.5 FlashGoogle | 89.5% | – | 60.4 | – |
| 11 |  Grok 4.6xAI | 89.4% | – | 58.7 | – |
| 12 |  Qwen3 7Alibaba | 89.3% | – | 58.9 | – |
| 13 | Gemini 3.6 FlashGoogle | 89.3% | – | 58.6 | – |
| 14 | Grok 4.5xAI | 89.2% | – | 54.2 | – |
| 15 | Claude Opus 4.6Anthropic | 89.1% | – | 62.3 | – |
| 16 |  GPT-5.6 SolOpenAI | 89.1% | – | 65.5 | – |
| 17 |  Muse Spark 1.1Meta | 88.7% | – | 56.4 | – |
| 18 | Qwen3 8Alibaba | 88.6% | – | 63.9 | – |
| 19 | Gemini 3 FlashGoogle | 88.6% | – | 58.5 | – |
| 20 | Muse Spark 1.2Meta | 88.3% | – | 61.3 | – |
| 21 | GPT-5.5OpenAI | 88.1% | – | 63.7 | – |
| 22 |  Kimi K3Moonshot AIopen | 88.0% | – | 62.7 | – |
| 23 | Claude Opus 4.1Anthropic | 87.9% | – | 58.1 | – |
| 24 | Qwen3.6 PlusAlibabaopen | 87.7% | – | 59.0 | – |
| 25 | Kimi K2.6Moonshot AIopen | 87.6% | kimi/kimi-k2.6 | 60.6 | – |
| 26 | Claude Sonnet 5Anthropic | 87.5% | – | 61.2 | – |
| 27 | GPT-5.4OpenAI | 87.5% | – | 61.3 | – |
| 28 | Claude Sonnet 4.5Anthropic | 87.4% | – | 52.8 | – |
| 29 | Claude Sonnet 4.6Anthropic | 87.3% | – | 60.3 | – |
| 30 | Muse SparkMeta | 87.3% | – | 65.4 | – |
| 31 | Claude Opus 4.5Anthropic | 87.3% | – | 60.6 | – |
| 32 |  DeepSeek V4 ProDeepSeekopen | 87.3% | – | 56.1 | – |
| 33 | Claude Opus 4.1Anthropic | 87.2% | – | 52.6 | – |
| 34 | Qwen3.5 PlusAlibaba | 87.2% | – | 57.2 | – |
| 35 |  MiniMax-M2MiniMaxopen | 87.0% | minimax/MiniMax-M2.1 | 55.1 | – |
| 36 |  GLM-5.1Zhipu AIopen | 86.9% | zai/glm-5.1 | 60.1 | – |
| 37 | GLM-5.3Zhipu AIopen | 86.8% | – | 59.9 | – |
| 38 | GLM-5.2Zhipu AIopen | 86.7% | zai/glm-5.2 | 57.4 | – |
| 39 | GPT-5.6 TerraOpenAI | 86.7% | – | 64.2 | – |
| 40 | GPT-5OpenAI | 86.5% | – | 54.5 | – |
| 41 | GPT-5.1OpenAI | 86.4% | – | 53.8 | – |
| 42 | Grok 4.20xAI | 86.3% | – | 54.3 | – |
| 43 | Gemini 3.1 Flash LiteGoogle | 86.2% | – | 54.3 | – |
| 44 | GPT-5.2OpenAI | 86.2% | – | 57.3 | – |
| 45 | DeepSeek V4 FlashDeepSeekopen | 86.2% | high, 0731 | 59.3 | – |
| 46 | Claude Opus 4Anthropic | 86.2% | – | 50.7 | – |
| 47 | GLM-5.3-FlashZhipu AIopen | 86.1% | – | 55.6 | – |
| 48 | GPT-5.6 LunaOpenAI | 86.0% | – | 56.5 | – |
| 49 | GLM-5Zhipu AIopen | 86.0% | – | 57.0 | – |
| 50 | Kimi K2.5Moonshot AIopen | 85.9% | – | 55.3 | – |
| 51 | Grok 4.3xAI | 85.8% | – | 50.1 | – |
| 52 | Gemini 3.5 Flash LiteGoogle | 85.8% | – | 44.0 | – |
| 53 |  Nemotron 3 Ultra 550B A55BNVIDIAopen | 85.8% | – | 58.3 | – |
| 54 | o3OpenAI | 85.6% | – | 52.9 | – |
| 55 | Claude Opus 4.5Anthropic | 85.6% | – | 58.0 | – |
| 56 | Inkling-SmallThinking Machinesopen | 85.6% | – | 55.0 | – |
| 57 | Grok 4xAI | 85.3% | 0709 | 57.2 | – |
| 58 | Qwen3 MaxAlibaba | 85.0% | – | 52.7 | – |
| 59 | DeepSeek V3.2DeepSeekopen | 84.9% | – | 55.9 | – |
| 60 | MiMo-V2.5-ProXiaomiopen | 84.6% | – | 59.0 | – |