Knowledge · %
MMLU
About 14,000 multiple-choice questions over 57 subjects, from elementary maths to law and medicine.
What it measures
Broad academic knowledge (usually 5-shot exact match). Data compiled by Epoch AI from papers, model reports and HELM.
Limitations
Saturated at the top; studies found a share of mislabelled questions; prompt format changes scores by several points.
Contamination risk: High
Public since 2020 and widely copied on the web.
Leaderboard · Subset / setting: compiled
96 models
| # | Model | Provider | Score | Rank | Setting | Source type | Date |
|---|---|---|---|---|---|---|---|
| 1 | GPT-4o (Nov 2024) | OpenAI | 88.1% | 1 | Exact match; 5-shot; effort: Nov 2024 | Paper / tech report via Epoch AI | Nov 20, 2024 |
| 2 | DeepSeek-V3OW | DeepSeek | 87.1% | 2 | Exact match; 5-shot | Paper / tech report via Epoch AI | Dec 26, 2024 |
| 3 | GPT-4 (Mar 2023) | OpenAI | 86.4% | 3 | Exact match; 5-shot; effort: Mar 2023 | Paper / tech report via Epoch AI | Mar 14, 2023 |
| 4 | Llama 3.3 70BOW | Meta | 86.3% | 4 | Exact match; 5-shot | Paper / tech report via Epoch AI | Dec 6, 2024 |
| 5 | Gemini 1.5 Pro (May 2024) | 85.9% | 5 | Exact match; 5-shot | Vendor-reported via Epoch AI | May 14, 2024 | |
| 6 | Qwen2.5-72BOW | Alibaba (Qwen) | 85.0% | 6 | Exact match; 5-shot | Paper / tech report via Epoch AI | Sep 19, 2024 |
| 7 | Phi-4OW | Microsoft Research | 84.8% | 7 | Exact match; 5-shot | Paper / tech report via Epoch AI | Dec 12, 2024 |
| 8 | Llama 3.1-405BOW | Meta | 84.4% | 8 | Exact match; 5-shot | Paper / tech report via Epoch AI | Jul 23, 2024 |
| 9 | GPT-4 | OpenAI | 83.9% | 9 | Exact match; 5-shot | Paper / tech report via Epoch AI | — |
| 10 | Gemini 1.5 Pro (Feb 2024) | 81.9% | 10 | Exact match; 5-shot | Vendor-reported via Epoch AI | Feb 15, 2024 | |
| 11 | GPT-4o mini | OpenAI | 81.8% | 11 | Exact match; 5-shot | Paper / tech report via Epoch AI | Jul 18, 2024 |
| 12 | Qwen2.5-14BOW | Alibaba (Qwen) | 79.9% | 12 | Exact match; 5-shot | Paper / tech report via Epoch AI | Sep 19, 2024 |
| 13 | Claude 2 | Anthropic | 78.5% | 13 | Exact match; 5-shot | Vendor-reported via Epoch AI | Jul 11, 2023 |
| 14 | DeepSeek-V2 (MoE-236B, May 2024)OW | DeepSeek | 78.4% | 14 | Exact match; 5-shot | Paper / tech report via Epoch AI | May 7, 2024 |
| 15 | phi-3-medium 14BOW | Microsoft | 77.9% | 15 | Exact match; 5-shot | Paper / tech report via Epoch AI | Apr 23, 2024 |
| 16 | Claude 1.3 | Anthropic | 77.0% | 16 | Exact match; 5-shot | Vendor-reported via Epoch AI | Apr 18, 2023 |
| 17 | Yi-34BOW | 01.AI | 76.3% | 17 | Exact match; 5-shot | Paper / tech report via Epoch AI | Nov 2, 2023 |
| 18 | phi-3-small 7.4BOW | Microsoft | 75.7% | 18 | Exact match; 5-shot | Paper / tech report via Epoch AI | Apr 23, 2024 |
| 19 | Qwen2.5-Coder-14BOW | Alibaba (Qwen) | 75.2% | 19 | Exact match | Paper / tech report via Epoch AI | Sep 18, 2024 |
| 20 | Claude Instant | Anthropic | 73.4% | 20 | Exact match; 5-shot | Vendor-reported via Epoch AI | — |
| 21 | Inflection-1 | Inflection AI | 72.7% | 21 | Exact match; 5-shot | Vendor-reported via Epoch AI | Jun 22, 2023 |
| 22 | GPT-3.5 Turbo (Nov 2023) | OpenAI | 71.4% | 22 | Exact match; 5-shot; effort: Nov 2023 | Paper / tech report via Epoch AI | Nov 6, 2023 |
| 23 | Falcon-180BOW | Technology Innovation Institute | 70.6% | 23 | Exact match; 5-shot | Paper / tech report via Epoch AI | Sep 6, 2023 |
| 24 | Mixtral 8x7BOW | Mistral AI | 70.6% | 23 | Exact match; 5-shot | Paper / tech report via Epoch AI | Dec 11, 2023 |
| 25 | davinci-002 | OpenAI | 70.0% | 25 | Exact match; 5-shot | Paper / tech report via Epoch AI | Mar 15, 2022 |
| 26 | PaLM (540B) | Google Research | 69.3% | 26 | Exact match; 5-shot | Paper / tech report via Epoch AI | Apr 4, 2022 |
| 27 | Llama 2-70BOW | Meta | 68.9% | 27 | Exact match; 5-shot | Paper / tech report via Epoch AI | Jul 18, 2023 |
| 28 | phi-3-mini 3.8BOW | Microsoft | 68.8% | 28 | Exact match; 5-shot | Paper / tech report via Epoch AI | Apr 23, 2024 |
| 29 | Stable Beluga 2OW | Stability AI | 68.6% | 29 | Exact match; 5-shot | Paper / tech report via Epoch AI | Jul 20, 2023 |
| 30 | GPT-3.5 Turbo | OpenAI | 68.5% | 30 | Exact match; 5-shot | Paper / tech report via Epoch AI | — |
| 31 | Yi-9BOW | 01.AI | 68.4% | 31 | Exact match; 5-shot | Paper / tech report via Epoch AI | Mar 1, 2024 |
| 32 | Qwen2.5-Coder (7B)OW | Alibaba (Qwen) | 68.0% | 32 | Exact match | Paper / tech report via Epoch AI | Sep 18, 2024 |
| 33 | Chinchilla | 67.5% | 34 | Exact match; 5-shot | Paper / tech report via Epoch AI | Mar 29, 2022 | |
| 34 | InternLM (104B) | Shanghai AI Lab (InternLM) | 67.2% | 35 | Exact match; 5-shot | Vendor-reported via Epoch AI | — |
| 35 | AquilaChat2 34B | — | 66.7% | 36 | Exact match; 5-shot | Paper / tech report via Epoch AI | — |
| 36 | Llama 3-8BOW | Meta | 66.5% | 37 | Exact match; 5-shot | Paper / tech report via Epoch AI | Apr 18, 2024 |
| 37 | Qwen-14BOW | Alibaba (Qwen) | 66.3% | 38 | Exact match; 5-shot | Paper / tech report via Epoch AI | Sep 28, 2023 |
| 38 | Gemma 7BOW | 64.3% | 40 | Exact match; 5-shot | Paper / tech report via Epoch AI | Feb 21, 2024 | |
| 39 | StarCoder 2 15BOW | Hugging Face | 64.1% | 41 | Exact match | Paper / tech report via Epoch AI | Feb 20, 2024 |
| 40 | LLaMA-65BOW | Meta | 63.4% | 43 | Exact match; 5-shot | Paper / tech report via Epoch AI | Feb 24, 2023 |
| 41 | Yi 6BOW | 01.AI | 63.2% | 44 | Exact match; 5-shot | Paper / tech report via Epoch AI | Nov 2, 2023 |
| 42 | Llama 2-34B | Meta | 62.6% | 45 | Exact match; 5-shot | Paper / tech report via Epoch AI | Jul 18, 2023 |
| 43 | Mistral 7B v0.2OW | Mistral AI | 62.5% | 46 | Exact match; 5-shot | Paper / tech report via Epoch AI | Dec 11, 2023 |
| 44 | Mistral 7B v0.1OW | Mistral AI | 62.5% | 46 | Exact match; 5-shot | Paper / tech report via Epoch AI | Sep 27, 2023 |
| 45 | Qwen2.5-Coder-32BOW | Alibaba (Qwen) | 61.2% | 48 | Exact match | Paper / tech report via Epoch AI | Sep 18, 2024 |
| 46 | DeepSeek-Coder-V2-Lite-Base | DeepSeek | 60.5% | 49 | Exact match | Paper / tech report via Epoch AI | Jun 13, 2024 |
| 47 | Gopher (280B) | 60.0% | 50 | Exact match; 5-shot | Paper / tech report via Epoch AI | Dec 8, 2021 | |
| 48 | Baichuan2-13BOW | Baichuan | 59.2% | 52 | Exact match; 5-shot | Paper / tech report via Epoch AI | Sep 6, 2023 |
| 49 | Nemotron-4 15B | NVIDIA | 58.7% | 53 | Exact match; 5-shot | Paper / tech report via Epoch AI | Feb 26, 2024 |
| 50 | Falcon 2 11BOW | Technology Innovation Institute | 58.4% | 54 | Exact match; 5-shot | Paper / tech report via Epoch AI | May 9, 2024 |
| 51 | LLaMA-33BOW | Meta | 57.8% | 56 | Exact match; 5-shot | Paper / tech report via Epoch AI | Feb 24, 2023 |
| 52 | internlm-chat-20b | Shanghai AI Lab (InternLM) | 57.4% | 57 | Exact match; 5-shot | Paper / tech report via Epoch AI | Sep 17, 2023 |
| 53 | Falcon-40BOW | Technology Innovation Institute | 56.9% | 58 | Exact match; 5-shot | Paper / tech report via Epoch AI | Mar 15, 2023 |
| 54 | Phi-2OW | Microsoft | 56.3% | 59 | Exact match; 5-shot | Paper / tech report via Epoch AI | Dec 12, 2023 |
| 55 | XVERSE-13B | — | 55.2% | 61 | Exact match; 5-shot | Paper / tech report via Epoch AI | — |
| 56 | Baichuan2-13B-Chat | Baichuan | 55.1% | 62 | Exact match; 0-shot | Paper / tech report via Epoch AI | Sep 6, 2023 |
| 57 | Llama 2-13BOW | Meta | 54.8% | 63 | Exact match; 5-shot | Paper / tech report via Epoch AI | Jul 18, 2023 |
| 58 | Baichuan 2-7BOW | Baichuan | 54.2% | 64 | Exact match; 5-shot | Paper / tech report via Epoch AI | Sep 20, 2023 |
| 59 | PaLM 62B | 53.7% | 65 | Exact match; 5-shot | Paper / tech report via Epoch AI | Apr 4, 2022 | |
| 60 | Qwen2.5-Coder (1.5B)OW | Alibaba (Qwen) | 53.6% | 66 | Exact match | Paper / tech report via Epoch AI | Sep 18, 2024 |
| 61 | Vicuna-13BOW | LMSYS | 52.0% | 67 | Exact match; 5-shot | Paper / tech report via Epoch AI | — |
| 62 | Baichuan 1-13BOW | Baichuan | 51.6% | 68 | Exact match; 5-shot | Paper / tech report via Epoch AI | Jul 11, 2023 |
| 63 | internlm-7b | Shanghai AI Lab (InternLM) | 51.0% | 69 | Exact match; 5-shot | Paper / tech report via Epoch AI | Jul 5, 2023 |
| 64 | INTELLECT-1 | Prime Intellect | 49.9% | 72 | Exact match; 5-shot | Paper / tech report via Epoch AI | Nov 29, 2024 |
| 65 | chatglm2-6bOW | Zhipu AI (Z.ai) | 47.9% | 73 | Exact match; 5-shot | Paper / tech report via Epoch AI | Jun 24, 2023 |
| 66 | MPT-30BOW | MosaicML | 47.9% | 73 | Exact match; 5-shot | Paper / tech report via Epoch AI | Jun 22, 2023 |
| 67 | ChatGLM 2-6B (base)* | Zhipu AI (Z.ai) | 47.9% | 75 | Exact match; 5-shot | Paper / tech report via Epoch AI | — |
| 68 | LLaMA-13BOW | Meta | 46.9% | 76 | Exact match; 5-shot | Paper / tech report via Epoch AI | Feb 24, 2023 |
| 69 | Llama 2-7BOW | Meta | 45.3% | 77 | Exact match; 5-shot | Paper / tech report via Epoch AI | Jul 18, 2023 |
| 70 | Qwen-7BOW | Alibaba (Qwen) | 45.0% | 79 | Exact match; 5-shot | Paper / tech report via Epoch AI | Sep 28, 2023 |
| 71 | Chinese-Alpaca-Plus-13B | Community (Chinese-LLaMA) | 43.9% | 80 | Exact match; 5-shot | Paper / tech report via Epoch AI | — |
| 72 | InstructGPT 175B | OpenAI | 43.9% | 80 | Exact match; 5-shot | Paper / tech report via Epoch AI | Jan 27, 2022 |
| 73 | Baichuan1-7BOW | Baichuan | 42.3% | 82 | Exact match; 5-shot | Paper / tech report via Epoch AI | Jun 1, 2023 |
| 74 | Gemma 2BOW | 42.3% | 82 | Exact match; 5-shot | Paper / tech report via Epoch AI | Feb 21, 2024 | |
| 75 | Qwen2.5-Coder-0.5B | Alibaba (Qwen) | 42.0% | 84 | Exact match | Paper / tech report via Epoch AI | Sep 18, 2024 |
| 76 | CodeQwen1.5-7BOW | Alibaba (Qwen) | 40.5% | 85 | Exact match | Paper / tech report via Epoch AI | Apr 15, 2024 |
| 77 | DeepSeek Coder 33BOW | DeepSeek | 39.4% | 86 | Exact match | Paper / tech report via Epoch AI | Nov 2, 2023 |
| 78 | StarCoder 2 7BOW | Hugging Face | 38.8% | 87 | Exact match | Paper / tech report via Epoch AI | Feb 20, 2024 |
| 79 | Phi-1.5OW | Microsoft | 37.6% | 88 | Exact match; 2-shot | Paper / tech report via Epoch AI | Sep 11, 2023 |
| 80 | StarCoder 2 3BOW | Hugging Face | 36.6% | 89 | Exact match | Paper / tech report via Epoch AI | Feb 22, 2024 |
| 81 | DeepSeek Coder 6.7BOW | DeepSeek | 36.4% | 90 | Exact match | Paper / tech report via Epoch AI | Nov 2, 2023 |
| 82 | XGen-7BOW | Salesforce | 36.3% | 91 | Exact match; 5-shot | Paper / tech report via Epoch AI | Jun 27, 2023 |
| 83 | LLaMA-7BOW | Meta | 35.2% | 92 | Exact match; 2-shot | Paper / tech report via Epoch AI | Feb 24, 2023 |
| 84 | Falcon-7BOW | Technology Innovation Institute | 35.0% | 94 | Exact match; 5-shot | Paper / tech report via Epoch AI | Apr 24, 2023 |
| 85 | GPT-NeoX 20BOW | EleutherAI | 33.6% | 95 | Exact match; 5-shot | Paper / tech report via Epoch AI | — |
| 86 | MPT-7BOW | MosaicML | 30.8% | 98 | Exact match; 5-shot | Paper / tech report via Epoch AI | May 5, 2023 |
| 87 | open_llama_7b | — | 29.9% | 99 | Exact match; 5-shot | Paper / tech report via Epoch AI | Jun 7, 2023 |
| 88 | Qwen-1_8B | Alibaba (Qwen) | 28.2% | 101 | Exact match; 5-shot | Paper / tech report via Epoch AI | Nov 30, 2023 |
| 89 | RedPajama-INCITE-7B-Base | — | 26.3% | 105 | Exact match; 5-shot | Paper / tech report via Epoch AI | May 4, 2023 |
| 90 | Cerebras-GPT-13BOW | Cerebras Systems | 26.2% | 106 | Exact match; 5-shot | Paper / tech report via Epoch AI | Mar 20, 2023 |
| 91 | Dolly 2.0-12bOW | Databricks | 26.2% | 106 | Exact match; 5-shot | Paper / tech report via Epoch AI | Apr 11, 2023 |
| 92 | Falcon-rw-1.3B | TII | 25.9% | 108 | Exact match; 2-shot | Paper / tech report via Epoch AI | — |
| 93 | DeepSeek Coder 1.3BOW | DeepSeek | 25.8% | 109 | Exact match | Paper / tech report via Epoch AI | Nov 2, 2023 |
| 94 | GPT-J-6BOW | EleutherAI | 25.7% | 111 | Exact match; 0-shot | Paper / tech report via Epoch AI | Aug 5, 2021 |
| 95 | PaLM 8B | 25.4% | 112 | Exact match; 5-shot | Paper / tech report via Epoch AI | — | |
| 96 | opt-13b | — | 25.1% | 114 | Exact match; 5-shot | Paper / tech report via Epoch AI | May 11, 2022 |
Leaderboard · Subset / setting: leaderboard
79 models
| # | Model | Provider | Score | Rank | Setting | Source type | Date |
|---|---|---|---|---|---|---|---|
| 1 | Claude 3.5 Sonnet (October 2024) | Anthropic | 87.3% | 1 | Exact match; 5-shot; effort: Oct 2024 | Official leaderboard via Epoch AI | Oct 22, 2024 |
| 2 | DeepSeek-V3OW | DeepSeek | 87.2% | 2 | Exact match; 5-shot | Official leaderboard via Epoch AI | Dec 26, 2024 |
| 3 | Gemini 1.5 Pro (Sept 2024) | 86.9% | 3 | Exact match; 5-shot | Official leaderboard via Epoch AI | Sep 24, 2024 | |
| 4 | Claude 3.5 Sonnet | Anthropic | 86.5% | 4 | Exact match; 5-shot; effort: Jun 2024 | Official leaderboard via Epoch AI | Jun 20, 2024 |
| 5 | Claude 3 Opus | Anthropic | 84.6% | 5 | Exact match; 5-shot | Official leaderboard via Epoch AI | Feb 29, 2024 |
| 6 | Llama 3.1-405BOW | Meta | 84.5% | 6 | Exact match; 5-shot | Official leaderboard via Epoch AI | Jul 23, 2024 |
| 7 | GPT-4o (Aug 2024) | OpenAI | 84.3% | 7 | Exact match; 5-shot; effort: Aug 2024 | Official leaderboard via Epoch AI | Aug 6, 2024 |
| 8 | GPT-4o (May 2024) | OpenAI | 84.2% | 8 | Exact match; 5-shot; effort: May 2024 | Official leaderboard via Epoch AI | May 13, 2024 |
| 9 | Qwen2.5-72BOW | Alibaba (Qwen) | 83.4% | 9 | Exact match; 5-shot | Official leaderboard via Epoch AI | Sep 19, 2024 |
| 10 | Gemini 1.5 Pro (Feb 2024) | 82.7% | 10 | Exact match; 5-shot | Official leaderboard via Epoch AI | Feb 15, 2024 | |
| 11 | GPT-4 (Jun 2023) | OpenAI | 82.4% | 11 | Exact match; 5-shot; effort: Jun 2023 | Official leaderboard via Epoch AI | Jun 13, 2023 |
| 12 | Qwen2-72BOW | Alibaba (Qwen) | 82.4% | 11 | Exact match; 5-shot | Official leaderboard via Epoch AI | Jun 7, 2024 |
| 13 | Amazon Nova Pro | Amazon | 82.0% | 13 | Exact match; 5-shot | Official leaderboard via Epoch AI | Dec 3, 2024 |
| 14 | GPT-4 Turbo (Apr 2024) | OpenAI | 81.3% | 14 | Exact match; 5-shot | Official leaderboard via Epoch AI | Apr 9, 2024 |
| 15 | Gemini 1.5 Pro (0409 preview) | 81.0% | 15 | Exact match; 5-shot | Official leaderboard via Epoch AI | — | |
| 16 | Llama 3.2 90BOW | Meta | 80.3% | 16 | Exact match; 5-shot | Official leaderboard via Epoch AI | Sep 24, 2024 |
| 17 | Llama 3.1-70BOW | Meta | 80.1% | 17 | Exact match; 5-shot | Official leaderboard via Epoch AI | Jul 23, 2024 |
| 18 | Mistral Large 2 (Jul 2024)OW | Mistral AI | 80.0% | 18 | Exact match; 5-shot | Official leaderboard via Epoch AI | Jul 24, 2024 |
| 19 | Gemini 2.0 Flash (Dec 2024) | 79.7% | 19 | Exact match; 5-shot | Official leaderboard via Epoch AI | Dec 11, 2024 | |
| 20 | GPT-4 Turbo (Nov 2023) | OpenAI | 79.6% | 20 | Exact match; 5-shot | Official leaderboard via Epoch AI | Nov 6, 2023 |
| 21 | Llama 3-70BOW | Meta | 79.3% | 21 | Exact match; 5-shot | Official leaderboard via Epoch AI | Apr 18, 2024 |
| 22 | Yi-Large | 01.AI | 79.3% | 21 | Exact match; 5-shot | Official leaderboard via Epoch AI | May 13, 2024 |
| 23 | Llama 3.3 70BOW | Meta | 79.1% | 23 | Exact match; 5-shot | Official leaderboard via Epoch AI | Dec 6, 2024 |
| 24 | PaLM-2 (Unicorn) | 78.6% | 24 | Exact match; 5-shot | Official leaderboard via Epoch AI | — | |
| 25 | Palmyra X V3 (72B) | 78.6% | 24 | Exact match; 5-shot | Official leaderboard via Epoch AI | — | |
| 26 | Jamba 1.5 LargeOW | AI21 Labs | 78.2% | 26 | Exact match; 5-shot | Official leaderboard via Epoch AI | — |
| 27 | Gemini 1.5 Flash (May 2024) | 77.9% | 27 | Exact match; 5-shot; effort: May 2024 | Official leaderboard via Epoch AI | May 23, 2024 | |
| 28 | Mixtral 8x22BOW | Mistral AI | 77.8% | 28 | Exact match; 5-shot | Official leaderboard via Epoch AI | Apr 17, 2024 |
| 29 | Solar Pro | — | 77.6% | 30 | Exact match; 5-shot | Official leaderboard via Epoch AI | — |
| 30 | phi-3-medium 14BOW | Microsoft | 77.5% | 31 | Exact match; 5-shot | Official leaderboard via Epoch AI | Apr 23, 2024 |
| 31 | Qwen1.5-72BOW | Alibaba (Qwen) | 77.4% | 32 | Exact match; 5-shot“Qwen1.5 (72B)” | Official leaderboard via Epoch AI | — |
| 32 | Amazon Nova Lite | Amazon | 77.0% | 33 | Exact match; 5-shot | Official leaderboard via Epoch AI | Dec 3, 2024 |
| 33 | Qwen1.5 Chat (110B) | Alibaba (Qwen) | 76.8% | 34 | Exact match; 5-shot | Official leaderboard via Epoch AI | — |
| 34 | GPT-4o mini | OpenAI | 76.7% | 35 | Exact match; 5-shot | Official leaderboard via Epoch AI | Jul 18, 2024 |
| 35 | Yi-34BOW | 01.AI | 76.2% | 36 | Exact match; 5-shot | Official leaderboard via Epoch AI | Nov 2, 2023 |
| 36 | Claude 3 Sonnet | Anthropic | 75.9% | 37 | Exact match; 5-shot | Official leaderboard via Epoch AI | Feb 29, 2024 |
| 37 | Gemma 2 27BOW | 75.7% | 38 | Exact match; 5-shot | Official leaderboard via Epoch AI | Jun 24, 2024 | |
| 38 | Phi-3 (7B) | Microsoft | 75.7% | 38 | Exact match; 5-shot | Official leaderboard via Epoch AI | — |
| 39 | Qwen1.5-32BOW | Alibaba (Qwen) | 74.4% | 40 | Exact match; 5-shot | Official leaderboard via Epoch AI | Feb 4, 2024 |
| 40 | Claude 3.5 Haiku | Anthropic | 74.3% | 41 | Exact match; 5-shot; effort: Oct 2024 | Official leaderboard via Epoch AI | Oct 22, 2024 |
| 41 | DBRXOW | Databricks | 74.1% | 42 | Exact match; 5-shot“DBRX Instruct” | Official leaderboard via Epoch AI | — |
| 42 | Gemini 1.5 Flash (Sep 2024) | 73.9% | 43 | Exact match; 5-shot | Official leaderboard via Epoch AI | Sep 24, 2024 | |
| 43 | Palmyra-X-004 | Writer | 73.9% | 43 | Exact match; 5-shot | Official leaderboard via Epoch AI | — |
| 44 | Claude 3 Haiku | Anthropic | 73.8% | 45 | Exact match; 5-shot | Official leaderboard via Epoch AI | Mar 7, 2024 |
| 45 | Claude 2.1 | Anthropic | 73.5% | 46 | Exact match; 5-shot | Official leaderboard via Epoch AI | Nov 21, 2023 |
| 46 | Qwen2.5-7BOW | Alibaba (Qwen) | 72.9% | 47 | Exact match; 5-shot | Official leaderboard via Epoch AI | Sep 19, 2024 |
| 47 | DeepSeek LLM Chat (67B) | DeepSeek | 72.5% | 48 | Exact match; 5-shot | Official leaderboard via Epoch AI | — |
| 48 | Gemma 2 9BOW | 72.1% | 49 | Exact match; 5-shot | Official leaderboard via Epoch AI | Jun 24, 2024 | |
| 49 | Mixtral (8x7B 32K seqlen) | Mistral AI | 71.7% | 50 | Exact match; 5-shot | Official leaderboard via Epoch AI | — |
| 50 | Amazon Nova Micro | Amazon | 70.8% | 51 | Exact match; 5-shot | Official leaderboard via Epoch AI | Dec 3, 2024 |
| 51 | Gemini 1.0 Pro | 70.0% | 52 | Exact match; 5-shot | Official leaderboard via Epoch AI | Dec 13, 2023 | |
| 52 | Jamba 1.5 MiniOW | AI21 Labs | 69.9% | 53 | Exact match; 5-shot | Official leaderboard via Epoch AI | — |
| 53 | Llama 2-70BOW | Meta | 69.5% | 54 | Exact match; 5-shot | Official leaderboard via Epoch AI | Jul 18, 2023 |
| 54 | Command R+OW | Cohere | 69.4% | 55 | Exact match; 5-shot | Official leaderboard via Epoch AI | Aug 30, 2024 |
| 55 | PaLM-2 (Bison) | 69.2% | 56 | Exact match; 5-shot | Official leaderboard via Epoch AI | — | |
| 56 | GPT-3.5 Turbo (Jun 2023) | OpenAI | 68.9% | 57 | Exact match; 5-shot; effort: June 2023 | Official leaderboard via Epoch AI | Jun 13, 2023 |
| 57 | Claude Instant | Anthropic | 68.8% | 58 | Exact match; 5-shot | Official leaderboard via Epoch AI | Aug 9, 2023 |
| 58 | Llama 3-8BOW | Meta | 68.8% | 58 | Exact match; 5-shot | Official leaderboard via Epoch AI | Apr 18, 2024 |
| 59 | Mistral Large | Mistral AI | 68.8% | 58 | Exact match; 5-shot | Official leaderboard via Epoch AI | Feb 26, 2024 |
| 60 | mistral-small-2402 | Mistral AI | 68.7% | 61 | Exact match; 5-shot | Official leaderboard via Epoch AI | Feb 26, 2024 |
| 61 | Qwen1.5-14BOW | Alibaba (Qwen) | 68.6% | 62 | Exact match; 5-shot | Official leaderboard via Epoch AI | Feb 4, 2024 |
| 62 | Arctic Instruct | Snowflake | 67.7% | 63 | Exact match; 5-shot | Official leaderboard via Epoch AI | — |
| 63 | GPT-3.5 Turbo (Jan 2024) | OpenAI | 67.3% | 64 | Exact match; 5-shot; effort: Jan 2024 | Official leaderboard via Epoch AI | Jan 25, 2024 |
| 64 | Gemma 7BOW | 66.1% | 65 | Exact match; 5-shot | Official leaderboard via Epoch AI | Feb 21, 2024 | |
| 65 | Jamba Instruct | AI21 Labs | 65.9% | 66 | Exact match; 5-shot | Official leaderboard via Epoch AI | — |
| 66 | Mistral NeMo (2402) | Mistral AI | 65.3% | 67 | Exact match; 5-shot | Official leaderboard via Epoch AI | — |
| 67 | c4ai-command-r-08-2024 | Cohere | 65.2% | 68 | Exact match; 5-shot | Official leaderboard via Epoch AI | Aug 30, 2024 |
| 68 | Yi 6BOW | 01.AI | 64.0% | 69 | Exact match; 5-shot | Official leaderboard via Epoch AI | Nov 2, 2023 |
| 69 | Qwen1.5-7BOW | Alibaba (Qwen) | 62.6% | 70 | Exact match; 5-shot | Official leaderboard via Epoch AI | Feb 4, 2024 |
| 70 | Palmyra X V2 (33B) | 62.1% | 71 | Exact match; 5-shot | Official leaderboard via Epoch AI | — | |
| 71 | Mistral 7B v0.3OW | Mistral AI | 59.9% | 72 | Exact match; 5-shot | Official leaderboard via Epoch AI | May 27, 2024 |
| 72 | Phi-2OW | Microsoft | 58.4% | 73 | Exact match; 5-shot | Official leaderboard via Epoch AI | Dec 12, 2023 |
| 73 | Mistral 7B v0.1OW | Mistral AI | 56.6% | 74 | Exact match; 5-shot | Official leaderboard via Epoch AI | Sep 27, 2023 |
| 74 | Llama 3.2 11BOW | Meta | 56.5% | 75 | Exact match; 5-shot | Official leaderboard via Epoch AI | Sep 24, 2024 |
| 75 | Llama 3.1-8BOW | Meta | 56.1% | 76 | Exact match; 5-shot | Official leaderboard via Epoch AI | Jul 23, 2024 |
| 76 | Llama 2-13BOW | Meta | 55.4% | 77 | Exact match; 5-shot | Official leaderboard via Epoch AI | Jul 18, 2023 |
| 77 | OLMo 1.7 (7B) | Ai2 | 53.8% | 78 | Exact match; 5-shot | Official leaderboard via Epoch AI | — |
| 78 | Llama 2-7BOW | Meta | 45.8% | 79 | Exact match; 5-shot | Official leaderboard via Epoch AI | Jul 18, 2023 |
| 79 | OLMo (7B) | Ai2 | 29.5% | 80 | Exact match; 5-shot | Official leaderboard via Epoch AI | — |
Best score over time
Each step is a new best result in our data, by evaluation date.
- 2023-06-1382.4%GPT-4 (Jun 2023)
- 2024-02-1582.7%Gemini 1.5 Pro (Feb 2024)
- 2024-02-2984.6%Claude 3 Opus
- 2024-06-2086.5%Claude 3.5 Sonnet
- 2024-09-2486.9%Gemini 1.5 Pro (Sept 2024)
- 2024-10-2287.3%Claude 3.5 Sonnet (October 2024)
Data sources & licences
| Source | Licence / terms | How we reuse it | Last fetched |
|---|---|---|---|
| Epoch AI Benchmarking Hub Epoch AI, 'Capabilities & benchmarking', https://epoch.ai/benchmarks (CC BY 4.0) | CC BY 4.0 | Open licence – full data with attribution | Oct 22, 2026 |
FAQ
What does MMLU measure?
About 14,000 multiple-choice questions over 57 subjects, from elementary maths to law and medicine. Broad academic knowledge (usually 5-shot exact match). Data compiled by Epoch AI from papers, model reports and HELM.
How reliable is MMLU?
Saturated at the top; studies found a share of mislabelled questions; prompt format changes scores by several points. Contamination risk: High – Public since 2020 and widely copied on the web.