Best LLM for knowledge: MMLU Pro ranking
- Claude Fable 5.1 leads
- 92.4%
- models measured
- 92
- lead over #2
- +0.8 pts
- median
- 86.0%
Leader on MMLU Pro · read October 20
As of October 20, 2026, Anthropic Claude Fable 5.1 leads knowledge with 92.4% on MMLU Pro, ahead of Anthropic Claude Opus 5 (91.6%) and Anthropic Claude Fable 5 (91.5%). This ranking is one public board, shown as published, and is not part of the Vibe Coding Index.
| Rank | Model | MMLU Pro | Coding evidence | Coding index | Output $/Mtok |
|---|---|---|---|---|---|
| 1 | Claude Fable 5.1 Anthropic | 92.4% | 89.1 | $50.00 | |
| 2 | Claude Opus 5 Anthropic | 91.6% | 81.8 | $25.00 | |
| 3 | Claude Fable 5 Anthropic | 91.5% | 82.1 | $50.00 | |
| 4 | Gemini 3.1 Pro Preview Google | 91.0% | 35.3 | $12.00 | |
| 5 | Gemini 3.8 Flash Google | 90.2% | 70.5 | $4.125 | |
| 6 | Gemini 3.7 Flash Google | 90.1% | 69.9 | $3.75 | |
| 7 | Claude Opus 4.7 Anthropic | 89.9% | 51.1 | $25.00 | |
| 8 | Claude Opus 4.8 Anthropic | 89.6% | 66.0 | $25.00 | |
| 9 | Gemini 3.5 Flash Google | 89.5% | 42.8 | $9.00 | |
| 10 | Grok 4.6 xAI | 89.4% | 76.4 | $6.00 | |
| 11 | Qwen3.7 Max Qwen | 89.3% | 45.1 | $4.425 | |
| 12 | Gemini 3.6 Flash Google | 89.3% | 48.3 | $3.75 | |
| 13 | Grok 4.5 xAI | 89.2% | 75.5 | $6.00 | |
| 14 | Claude Opus 4.6 Anthropic | 89.1% | 66.9 | $25.00 | |
| 15 | GPT-5.6 Sol OpenAI | 89.1% | 76.7 | $20.00 | |
| 16 | Muse Spark 1.1 Meta | 88.7% | 48.9 | $4.25 | |
| 17 | Qwen3.8 Max (0902) Qwen | 88.6% | 63.5 | $6.00 | |
| 18 | Gemini 3 Flash Preview Google | 88.6% | 75.8 | $3.00 | |
| 19 | Muse Spark 1.2 Meta | 88.3% | 65.1 | $4.25 | |
| 20 | GPT-5.5 OpenAI | 88.1% | 67.8 | $30.00 | |
| 21 | Kimi K3 Moonshot AI | 88.0% | 61.8 | $11.936 | |
| 22 | Claude Opus 4.1 Anthropic | 87.9% | 27.2 | $75.00 | |
| 23 | Qwen3.6 Plus Qwen | 87.7% | 37.4 | $1.95 | |
| 24 | Kimi K2.6 Moonshot AI | 87.6% | 44.2 | $3.41 | |
| 25 | Claude Sonnet 5 Anthropic | 87.5% | 63.5 | $10.00 | |
| 26 | GPT-5.4 OpenAI | 87.5% | 37.7 | $15.00 | |
| 27 | Claude Sonnet 4.5 Anthropic | 87.4% | 59.1 | $15.00 | |
| 28 | Claude Sonnet 4.6 Anthropic | 87.3% | 57.8 | $15.00 | |
| 29 | Claude Opus 4.5 Anthropic | 87.3% | 66.8 | $25.00 | |
| 30 | DeepSeek V4 Pro 0423 DeepSeek | 87.3% | 35.3 | $1.9105 | |
| 31 | 87.2% | Not benchmarked | $1.80 | ||
| 32 | MiniMax M2.1 MiniMax | 87.0% | Not benchmarked | $1.20 | |
| 33 | DeepSeek V4 Pro 0813 DeepSeek | 87.0% | 54.6 | $1.98 | |
| 34 | GLM 5.1 Z.ai | 86.9% | 44.0 | $3.036 | |
| 35 | GLM 5.3 Z.ai | 86.8% | 73.1 | $1.40 | |
| 36 | GLM 5.2 Z.ai | 86.7% | 57.7 | $2.42 | |
| 37 | GPT-5.6 Terra OpenAI | 86.7% | 69.0 | $12.00 | |
| 38 | GPT-5 OpenAI | 86.5% | 72.2 | $10.00 | |
| 39 | GPT-5.1 OpenAI | 86.4% | 53.3 | $10.00 | |
| 40 | T Inkling Thinking Machines | 86.3% | 49.7 | $4.05 | |
| 41 | Grok 4.20 xAI | 86.3% | Not benchmarked | $2.50 | |
| 42 | 86.2% | 8.1 | $1.50 | ||
| 43 | GPT-5.2 OpenAI | 86.2% | 58.9 | $14.00 | |
| 44 | DeepSeek V4 Flash 0731 DeepSeek | 86.2% | Not benchmarked | $1.32 | |
| 45 | GLM 5.3 Flash Z.ai | 86.1% | 71.0 | $0.16 | |
| 46 | GPT-5.6 Luna OpenAI | 86.0% | 61.5 | $1.20 | |
| 47 | GLM 5 Z.ai | 86.0% | 59.7 | $1.92 | |
| 48 | Kimi K2.5 Moonshot AI | 85.9% | 58.5 | $2.50 | |
| 49 | Gemini 3.5 Flash Lite Google | 85.8% | 33.6 | $2.50 | |
| 50 | Grok 4.3 xAI | 85.8% | 22.5 | $2.50 | |
| 51 | Nemotron 3 Ultra NVIDIA | 85.8% | Not benchmarked | $2.40 | |
| 52 | o3 OpenAI | 85.6% | 70.7 | $8.00 | |
| 53 | T Inkling Small Thinking Machines | 85.6% | 29.9 | $1.20 | |
| 54 | DeepSeek V3.2 DeepSeek | 84.9% | 54.4 | $0.42 | |
| 55 | MiMo-V2.5-Pro Xiaomi | 84.6% | 40.0 | $0.87 | |
| 56 | GPT-5.4 Mini OpenAI | 84.5% | 28.2 | $4.50 | |
| 57 | Qwen3.8 27B Qwen | 84.3% | 73.6 | $2.20 | |
| 58 | MiniMax M3 MiniMax | 84.2% | 57.8 | $1.20 | |
| 59 | Qwen3.5-Flash Qwen | 84.1% | 6.2 | $0.26 | |
| 60 | o1 OpenAI | 83.5% | 61.7 | $60.00 | |
| 61 | MiMo-V2.5 Xiaomi | 82.9% | 34.0 | $0.28 | |
| 62 | GLM 4.7 Z.ai | 82.7% | 33.6 | $1.98 | |
| 63 | GPT-5 Mini OpenAI | 82.2% | 59.8 | $2.00 | |
| 64 | GLM 4.6 Z.ai | 82.2% | 52.2 | $2.20 | |
| 65 | Ling 3.0 Flash inclusionAI | 82.0% | Not benchmarked | $0.063 | |
| 66 | GLM 4.5 Z.ai | 81.2% | 64.2 | $2.20 | |
| 67 | Kimi K2 Thinking Moonshot AI | 81.1% | 63.4 | $2.50 | |
| 68 | o4 Mini OpenAI | 80.6% | 68.6 | $4.40 | |
| 69 | GPT-4.1 OpenAI | 80.5% | 46.5 | $8.00 | |
| 70 | MiniMax M2.7 MiniMax | 80.4% | 28.4 | $1.20 | |
| 71 | MiniMax M2.5 MiniMax | 80.1% | 59.4 | $0.95 | |
| 72 | Mistral Large 3 2512 Mistral AI | 79.8% | Not benchmarked | $1.50 | |
| 73 | DeepSeek V3 0324 DeepSeek | 79.5% | 42.0 | $1.14 | |
| 74 | Llama 4 Maverick Meta Llama | 79.4% | 18.1 | $0.6525 | |
| 75 | Kimi K2 0711 Moonshot AI | 79.4% | 64.7 | $2.30 | |
| 76 | gpt-oss-120b OpenAI | 79.2% | 34.5 | $0.60 | |
| 77 | Gemini 2.5 Flash Lite Google | 79.1% | Not benchmarked | $0.40 | |
| 78 | Claude Haiku 4.5 Anthropic | 78.7% | 49.8 | $5.00 | |
| 79 | o3 Mini OpenAI | 78.7% | 52.1 | $4.40 | |
| 80 | GPT-4.1 Mini OpenAI | 77.2% | 28.5 | $1.60 | |
| 81 | GPT-5.4 Nano OpenAI | 77.2% | Not benchmarked | $1.25 | |
| 82 | GPT-5 Nano OpenAI | 76.1% | 34.8 | $0.40 | |
| 83 | Mistral Medium 3.5 Mistral AI | 75.3% | 9.2 | $7.50 | |
| 84 | Mistral Medium 3.1 Mistral AI | 75.3% | Not benchmarked | $2.00 | |
| 85 | GPT-4o (2024-08-06) OpenAI | 74.1% | 23.1 | $10.00 | |
| 86 | GPT-4o (2024-11-20) OpenAI | 72.6% | 18.2 | $10.00 | |
| 87 | gpt-oss-20b OpenAI | 71.6% | Not benchmarked | $0.09 | |
| 88 | Mistral Large Mistral AI | 69.7% | Not benchmarked | $6.00 | |
| 89 | Llama 4 Scout Meta Llama | 69.6% | 9.1 | $0.30 | |
| 90 | Command A Cohere | 69.2% | Not benchmarked | $10.00 | |
| 91 | GPT-4.1 Nano OpenAI | 63.5% | 8.9 | $0.40 | |
| 92 | GPT-4o-mini OpenAI | 62.7% | Not benchmarked | $0.60 |
What this measures
How much does the model know across fourteen academic and professional subjects?
MMLU Pro is over 12,000 multiple-choice questions across 14 subjects, from law and history to physics, each with ten options instead of four. The number is the share answered correctly, as run by Vals AI.
Limit: The top of the board is close together: when we read it on October 15, 2026, the five leading models all scored above 90%.
Read from MMLU Pro (Vals AI) on October 20, 2026; the board says it was last updated September 1, 2026. Each model appears once with its best published setting; rows that match no model in our catalog are left out. MMLU Pro benchmark page