Knowledge
MMLU Leaderboard
MMLU (Massive Multitask Language Understanding) spans 57 subjects from history to law to medicine as multiple-choice questions. It is the long-standing default for broad knowledge and remains the most widely-reported general benchmark.
Source: epoch76 open models ranked+60 proprietaryData through Feb 2025
Open models ranked on MMLU
# shows rank among open models / rank overall (including proprietary).
| # | Model | Score |
|---|---|---|
| 1 / 3 | DeepSeek v3 · 684.5B | 87.2% |
| 2 / 7 | Llama 3.3 70B Instruct · 70.6B | 86.3% |
| 3 / 9 | Qwen2.5 72B Instruct · 72.7B | 85.3% |
| 4 / 10 | Qwen2.5 72B · 72.7B | 85.0% |
| 5 / 11 | Phi 4 · 14.7B | 84.8% |
| 6 / 13 | Llama 3.1 405B Instruct · 405.9B | 84.5% |
| 7 / 14 | Llama 3.1 405B · 405.9B | 84.4% |
| 8 / 19 | Qwen2 72B Instruct · 72.7B | 82.4% |
| 9 / 23 | Llama 3.2 90B Vision Instruct · 88.6B | 80.3% |
| 10 / 24 | Llama 3.1 70B Instruct · 70.6B | 80.1% |
| 11 / 26 | Qwen2.5 14B Instruct · 14.8B | 79.9% |
| 12 / 29 | Meta Llama 3 70B Instruct · 70.6B | 79.3% |
| 13 / 31 | Qwen2.5 Coder 32B · 32.8B | 79.1% |
| 14 / 33 | DeepSeek v2 · 235.7B | 78.4% |
| 15 / 37 | Mixtral 8x22B v0.1 · 140.6B | 77.8% |
| 16 / 40 | Yi 34B · 34.4B | 76.3% |
| 17 / 42 | Gemma 2 27B IT · 27.2B | 75.7% |
| 18 / 43 | Phi 3 Small 8k Instruct · 7.4B | 75.7% |
| 19 / 44 | Qwen2.5 Coder 14B · 14.8B | 75.2% |
| 20 / 45 | Qwen1.5 32B · 32.5B | 74.4% |
| 21 / 50 | Yi 34B Chat · 34.4B | 73.5% |
| 22 / 53 | Qwen2.5 7B Instruct · 7.6B | 72.9% |
| 23 / 55 | Gemma 2 9B IT · 9.2B | 72.1% |
| 24 / 58 | Falcon 180B · 180B | 70.6% |
| 25 / 59 | Mixtral 8x7B v0.1 · 46.7B | 70.6% |
| 26 / 62 | Llama 2 70B HF · 69.0B | 69.9% |
| 27 / 66 | Meta Llama 3 8B Instruct · 8.0B | 68.8% |
| 28 / 68 | Phi 3 Mini 4k Instruct · 3.8B | 68.8% |
| 29 / 70 | Qwen1.5 14B · 14.2B | 68.6% |
| 30 / 71 | StableBeluga2 · 70B | 68.6% |
| 31 / 72 | Yi 9B · 8.8B | 68.4% |
| 32 / 73 | Qwen2.5 Coder 7B · 7.6B | 68.0% |
| 33 / 76 | Qwen 14B · 14.2B | 66.3% |
| 34 / 77 | Gemma 7B · 8.5B | 66.1% |
| 35 / 79 | Qwen 14B Chat · 14.2B | 65.0% |
| 36 / 80 | Starcoder2 15B · 16.0B | 64.1% |
| 37 / 81 | Yi 6B · 6.1B | 64.0% |
| 38 / 84 | Qwen1.5 7B · 7.7B | 62.6% |
| 39 / 85 | Mistral 7B Instruct v0.2 · 7.2B | 62.5% |
| 40 / 86 | Mistral 7B v0.1 · 7B | 62.5% |
| 41 / 87 | Yi 6B Chat · 6.1B | 61.0% |
| 42 / 88 | DeepSeek Coder v2 Lite Base · 15.7B | 60.5% |
| 43 / 90 | Llama 2 70B Chat HF · 69.0B | 59.9% |
| 44 / 91 | Mistral 7B Instruct v0.3 · 7.2B | 59.9% |
| 45 / 92 | Baichuan2 13B Base · 13B | 59.2% |
| 46 / 95 | Falcon 11B · 11.1B | 58.4% |
| 47 / 96 | Phi 2 · 2.8B | 58.4% |
| 48 / 97 | Internlm Chat 20B · 20B | 57.4% |
| 49 / 98 | Falcon 40B · 41.8B | 56.9% |
| 50 / 99 | Llama 3.2 11B Vision Instruct · 10.7B | 56.5% |
| 51 / 100 | Llama 3.1 8B Instruct · 8.0B | 56.1% |
| 52 / 101 | Llama 2 13B HF · 13.0B | 55.6% |
| 53 / 102 | Baichuan2 13B Chat · 13B | 55.1% |
| 54 / 103 | Baichuan2 7B Base · 7B | 54.2% |
| 55 / 105 | Qwen2.5 Coder 1.5B · 1.5B | 53.6% |
| 56 / 106 | Baichuan 13B Base · 13B | 51.6% |
| 57 / 107 | Internlm 7B · 7B | 51.0% |
| 58 / 108 | Llama 2 13B Chat HF · 13.0B | 50.9% |
| 59 / 109 | INTELLECT 1 Instruct · 10.2B | 49.9% |
| 60 / 110 | Chatglm2 6B · 6B | 47.9% |
| 61 / 113 | Llama 2 7B HF · 6.7B | 45.8% |
| 62 / 114 | Qwen 7B · 7.7B | 45.0% |
| 63 / 116 | Baichuan 7B · 7B | 42.3% |
| 64 / 117 | Gemma 2B · 2.5B | 42.3% |
| 65 / 118 | Qwen2.5 Coder 0.5B · 494M | 42.0% |
| 66 / 119 | CodeQwen1.5 7B · 7.3B | 40.5% |
| 67 / 121 | Starcoder2 7B · 7.2B | 38.8% |
| 68 / 122 | Phi 1 5 · 1.4B | 37.6% |
| 69 / 123 | Starcoder2 3B · 3.0B | 36.6% |
| 70 / 125 | Xgen 7B 8k Base · 7B | 36.3% |
| 71 / 126 | Llama 7B · 6.7B | 35.6% |
| 72 / 127 | Falcon 7B · 7.2B | 35.0% |
| 73 / 130 | Qwen 1 8B · 1.8B | 28.2% |
| 74 / 132 | Cerebras GPT 13B · 13B | 26.2% |
| 75 / 134 | Deepseek Coder 1.3B Base · 1.3B | 25.8% |
| 76 / 135 | GPT J 6B · 6B | 25.7% |
Score vs model size
Which models give the most quality for their size — the ones worth running locally.
- Qwen2.5 Coder 0.5B, 494M, score 42.0% — on the efficiency frontier (best score at its size or smaller).
- Qwen2.5 Coder 1.5B, 2B, score 53.6% — on the efficiency frontier (best score at its size or smaller).
- Phi 2, 3B, score 58.4% — on the efficiency frontier (best score at its size or smaller).
- Phi 3 Mini 4k Instruct, 4B, score 68.8% — on the efficiency frontier (best score at its size or smaller).
- Phi 3 Small 8k Instruct, 7B, score 75.7% — on the efficiency frontier (best score at its size or smaller).
- Phi 4, 15B, score 84.8% — on the efficiency frontier (best score at its size or smaller).
- Llama 3.3 70B Instruct, 71B, score 86.3% — on the efficiency frontier (best score at its size or smaller).
- DeepSeek v3, 685B, score 87.2% — on the efficiency frontier (best score at its size or smaller).
MMLU: frequently asked questions
- What is the best open LLM on MMLU?
- DeepSeek v3 is the top open model on MMLU, scoring 87.2%. Among all models tested — including proprietary ones — it ranks #3. The top model overall is GPT 4o (Nov 20, 2024) (OpenAI) at 88.1%.
- What's the best MMLU model you can run on a 24 GB GPU?
- Phi 4 is the highest-scoring open model that fits in 24 GB at 4-bit quantization (about 8 GB), scoring 84.8% on MMLU.
- What's the best MMLU model you can run on a 12 GB GPU?
- Phi 4 is the highest-scoring open model that fits in 12 GB at 4-bit quantization (about 8 GB), scoring 84.8% on MMLU.
- Can open models match proprietary models on MMLU?
- Not quite on MMLU: the strongest proprietary model (GPT 4o (Nov 20, 2024)) scores 88.1%, ahead of the best open model (DeepSeek v3) at 87.2% — but you can run the open one yourself.
Scores aggregated from epoch. llmrun does not run this benchmark — see the source for methodology, or the about benchmarks for what it measures.