knowledge benchmark · included in ECI
MMLU
MMLU benchmark (score column: EM).
Results
68
Random baseline
25.0%
Score ceiling
100%
Released
7 Sep 2020
Score by model release date
Best score by organisation
Leaderboard
| # | Model | Score | Relative | Setting | Released | Run | Evidence |
|---|---|---|---|---|---|---|---|
| 1 | DeepSeek V3 Open source DeepSeek | 87.2% | — | 26 Dec 2024 | — | Stanford CRFM Leaderboard ↗ | |
| 2 | Llama 3.3 70B Open source Meta AI | 86.3% | — | 6 Dec 2024 | — | Phi-4 Technical Report ↗ | |
| 3 | Qwen2.5-72B Open source Alibaba | 85.3% | — | 19 Sep 2024 | — | Phi-4 Technical Report ↗ | |
| 4 | Phi 4 Open source Microsoft | 84.8% | — | 12 Dec 2024 | — | Phi-4 Technical Report ↗ | |
| 5 | Llama 3.1-405B Open source Meta AI | 84.5% | — | 23 Jul 2024 | — | Stanford CRFM Leaderboard ↗ | |
| 6 | Qwen2-72B Open source Alibaba | 82.4% | — | 7 Jun 2024 | — | Stanford CRFM Leaderboard ↗ | |
| 7 | Llama 3.2 90B Open source Meta AI | 80.3% | — | 24 Sep 2024 | — | Stanford CRFM Leaderboard ↗ | |
| 8 | Llama 3.1-70B Open source Meta AI | 80.1% | — | 23 Jul 2024 | — | Stanford CRFM Leaderboard ↗ | |
| 9 | Mistral Large 2 (Jul 2024) Open source Mistral AI | 80.0% | — | 24 Jul 2024 | — | Stanford CRFM Leaderboard ↗ | |
| 10 | Qwen2.5-14B Open source Alibaba | 79.9% | — | 19 Sep 2024 | — | Phi-4 Technical Report ↗ | |
| 11 | Llama 3-70B Open source Meta AI | 79.3% | — | 18 Apr 2024 | — | Stanford CRFM Leaderboard ↗ | |
| 12 | Qwen2.5-Coder-32B Open source Alibaba | 79.1% | — | 18 Sep 2024 | — | Qwen2.5-Coder Technical Rep… ↗ | |
| 13 | DeepSeek-V2 (MoE-236B, May 2024) Open source DeepSeek | 78.4% | — | 7 May 2024 | — | DeepSeek-V3 Technical Report ↗ | |
| 14 | phi-3-medium 14B Open source Microsoft | 78.0% | — | 23 Apr 2024 | — | Phi-3 Technical Report: A H… ↗ | |
| 15 | Mixtral 8x22B Open source Mistral AI | 77.8% | — | 17 Apr 2024 | — | Stanford CRFM Leaderboard ↗ | |
| 16 | Yi-34B Open source 01.AI | 76.3% | — | 2 Nov 2023 | — | Yi: Open Foundation Models … ↗ | |
| 17 | Gemma 2 27B Open source Google | 75.7% | — | 24 Jun 2024 | — | Stanford CRFM Leaderboard ↗ | |
| 18 | phi-3-small 7.4B Open source Microsoft | 75.7% | — | 23 Apr 2024 | — | Phi-3 Technical Report: A H… ↗ | |
| 19 | Qwen1.5-32B Open source Alibaba | 74.4% | — | 3 Apr 2024 | — | Stanford CRFM Leaderboard ↗ | |
| 20 | Qwen2.5-7B Open source Alibaba | 72.9% | — | 19 Sep 2024 | — | Stanford CRFM Leaderboard ↗ | |
| 21 | Gemma 2 9B Open source Google DeepMind | 72.1% | — | 24 Jun 2024 | — | Stanford CRFM Leaderboard ↗ | |
| 22 | Mixtral 8x7B Open source Mistral AI | 70.6% | — | 11 Dec 2023 | — | Mixtral of Experts ↗ | |
| 23 | Falcon-180B Open source Technology Innovation Institute | 70.6% | — | 6 Sep 2023 | — | The Falcon Series of Open L… ↗ | |
| 24 | Llama 2-70B Open source Meta AI | 69.9% | — | 18 Jul 2023 | — | Mixtral of Experts ↗ | |
| 25 | Command R+ Open source Cohere,Cohere Labs (formerly Cohere for AI) | 69.4% | — | 30 Aug 2024 | — | Stanford CRFM Leaderboard ↗ | |
| 26 | phi-3-mini 3.8B Open source Microsoft | 68.8% | — | 23 Apr 2024 | — | Phi-3 Technical Report: A H… ↗ | |
| 27 | Llama 3-8B Open source Meta AI | 68.8% | — | 18 Apr 2024 | — | Stanford CRFM Leaderboard ↗ | |
| 28 | Qwen1.5-14B Open source Alibaba | 68.6% | — | 4 Feb 2024 | — | Stanford CRFM Leaderboard ↗ | |
| 29 | Stable Beluga 2 Open source Stability AI | 68.6% | — | 20 Jul 2023 | — | Qwen Technical Report ↗ | |
| 30 | Qwen2.5-Coder (7B) Open source Alibaba | 68.0% | — | 18 Sep 2024 | — | Qwen2.5-Coder Technical Rep… ↗ | |
| 31 | Qwen-14B Open source Alibaba | 66.3% | — | 24 Sep 2023 | — | Nemotron-4 15B Technical Re… ↗ | |
| 32 | Gemma 7B Open source Google DeepMind | 66.1% | — | 21 Feb 2024 | — | Stanford CRFM Leaderboard ↗ | |
| 33 | StarCoder 2 15B Open source Hugging Face,ServiceNow,Nvidia,BigCode | 64.1% | — | 20 Feb 2024 | — | Qwen2.5-Coder Technical Rep… ↗ | |
| 34 | Yi 6B Open source 01.AI | 64.0% | — | 22 Nov 2023 | — | Stanford CRFM Leaderboard ↗ | |
| 35 | LLaMA-65B Open source Meta AI | 63.4% | — | 24 Feb 2023 | — | Llama 2: Open Foundation an… ↗ | |
| 36 | Qwen1.5-7B Open source Alibaba | 62.6% | — | 4 Feb 2024 | — | Stanford CRFM Leaderboard ↗ | |
| 37 | Mistral 7B v0.2 Open source Mistral AI | 62.5% | — | 11 Dec 2023 | — | Gemma: Open Models Based on… ↗ | |
| 38 | Mistral 7B v0.1 Open source Mistral AI | 62.5% | — | 27 Sep 2023 | — | Mixtral of Experts ↗ | |
| 39 | Mistral 7B v0.3 Open source Mistral AI | 59.9% | — | 27 May 2024 | — | Stanford CRFM Leaderboard ↗ | |
| 40 | Baichuan2-13B Open source Baichuan | 59.2% | — | 6 Sep 2023 | — | Nemotron-4 15B Technical Re… ↗ | |
| 41 | LLaMA-33B Open source Meta AI | 58.7% | — | 24 Feb 2023 | — | Qwen Technical Report ↗ | |
| 42 | Falcon 2 11B Open source Technology Innovation Institute | 58.4% | — | 9 May 2024 | — | Falcon2-11B Technical Report ↗ | |
| 43 | Phi-2 Open source Microsoft | 58.4% | — | 12 Dec 2023 | — | Stanford CRFM Leaderboard ↗ | |
| 44 | Falcon-40B Open source Technology Innovation Institute | 56.9% | — | 25 May 2023 | — | Falcon2-11B Technical Repor… ↗ | |
| 45 | Llama 3.2 11B Open source Meta AI | 56.5% | — | 24 Sep 2024 | — | Stanford CRFM Leaderboard ↗ | |
| 46 | Llama 3.1-8B Open source Meta AI | 56.1% | — | 23 Jul 2024 | — | Stanford CRFM Leaderboard ↗ | |
| 47 | Llama 2-13B Open source Meta AI | 55.6% | — | 18 Jul 2023 | — | Mixtral of Experts ↗ | |
| 48 | Baichuan 2-7B Open source Baichuan | 54.2% | — | 20 Sep 2023 | — | Baichuan 2: Open Large-scal… ↗ | |
| 49 | Qwen2.5-Coder (1.5B) Open source Alibaba | 53.6% | — | 18 Sep 2024 | — | Qwen2.5-Coder Technical Rep… ↗ | |
| 50 | Baichuan 1-13B Open source Baichuan | 51.6% | — | 11 Jul 2023 | — | Baichuan 2: Open Large-scal… ↗ | |
| 51 | MPT-30B Open source MosaicML | 47.9% | — | 22 Jun 2023 | — | Qwen Technical Report ↗ | |
| 52 | LLaMA-13B Open source Meta AI | 47.7% | — | 24 Feb 2023 | — | Qwen Technical Report ↗ | |
| 53 | Llama 2-7B Open source Meta AI | 45.8% | — | 18 Jul 2023 | — | Stanford CRFM Leaderboard ↗ | |
| 54 | Qwen-7B Open source Alibaba | 45.0% | — | 28 Sep 2023 | — | Qwen Technical Report ↗ | |
| 55 | Gemma 2B Open source Google DeepMind | 42.3% | — | 21 Feb 2024 | — | Gemma: Open Models Based on… ↗ | |
| 56 | Baichuan1-7B Open source Baichuan | 42.3% | — | 1 Jun 2023 | — | Baichuan 2: Open Large-scal… ↗ | |
| 57 | DeepSeek Coder 33B Open source DeepSeek,Peking University | 39.4% | — | 2 Nov 2023 | — | Qwen2.5-Coder Technical Rep… ↗ | |
| 58 | StarCoder 2 7B Open source Hugging Face,ServiceNow,Nvidia,BigCode | 38.8% | — | 20 Feb 2024 | — | Qwen2.5-Coder Technical Rep… ↗ | |
| 59 | Phi-1.5 Open source Microsoft | 37.6% | 5 | 11 Sep 2023 | — | Textbooks Are All You Need … ↗ | |
| 60 | StarCoder 2 3B Open source Hugging Face,ServiceNow,Nvidia,BigCode | 36.6% | — | 22 Feb 2024 | — | Qwen2.5-Coder Technical Rep… ↗ | |
| 61 | DeepSeek Coder 6.7B Open source DeepSeek,Peking University | 36.4% | — | 2 Nov 2023 | — | Qwen2.5-Coder Technical Rep… ↗ | |
| 62 | XGen-7B Open source Salesforce | 36.3% | — | 27 Jun 2023 | — | XGen-7B Technical Report ↗ | |
| 63 | LLaMA-7B Open source Meta AI | 35.6% | — | 24 Feb 2023 | — | Qwen Technical Report ↗ | |
| 64 | Falcon-7B Open source Technology Innovation Institute | 35.0% | — | 24 Apr 2023 | — | Falcon2-11B Technical Report ↗ | |
| 65 | MPT-7B Open source MosaicML | 30.8% | — | 5 May 2023 | — | Qwen Technical Report ↗ | |
| 66 | Dolly 2.0-12b Open source Databricks | 26.2% | — | 11 Apr 2023 | — | XGen-7B Technical Report ↗ | |
| 67 | Cerebras-GPT-13B Open source Cerebras Systems | 26.2% | — | 20 Mar 2023 | — | XGen-7B Technical Report ↗ | |
| 68 | DeepSeek Coder 1.3B Open source DeepSeek,Peking University | 25.8% | — | 2 Nov 2023 | — | Qwen2.5-Coder Technical Rep… ↗ |
Caveats: settings (reasoning effort, agent scaffold) differ between rows and materially affect scores; the best reported setting per model is shown. Source: Epoch AI, CC BY 4.0.