Knowledge · %
MMLU-Pro
Harder successor to MMLU: about 12,000 reasoning-heavy questions in 14 disciplines with 10 answer options.
Active Contamination risk: Medium 247 results · 243 models Updated Oct 22, 2026 Newest result: 2026-03-11 · stale Official leaderboard ↗
What it measures
Knowledge plus reasoning, usually with chain-of-thought.
Limitations
The official leaderboard mixes TIGER-Lab runs with self-reported numbers; we label each row.
Contamination risk: Medium
Public test set since 2024.
Leaderboard
Top 100 of 243
| # | Model | Provider | Score | Rank | Setting | Source type | Date |
|---|---|---|---|---|---|---|---|
| 1 | Gemini 3.1 Pro | 91.2% | 1 | Overall accuracy; data source: TIGER-Lab“Gemini-3.1-Pro” | Official leaderboard | Mar 11, 2026 | |
| 2 | Gemini 3 Pro | 90.1% | 2 | Overall accuracy; data source: Self-Reported“Gemini-3-Pro(11/25)” | Vendor-reported | Mar 11, 2026 | |
| 3 | GPT-o1 | OpenAI | 89.3% | 3 | Overall accuracy; data source: Self-Reported | Vendor-reported | Mar 11, 2026 |
| 4 | Claude-4.6-Opus(Thinking) | Anthropic | 89.1% | 4 | Overall accuracy; data source: Self-Reported | Vendor-reported | Mar 11, 2026 |
| 5 | 25) | 88.6% | 5 | Overall accuracy; data source: Self-Reported“Gemini-3-Flash(12/25)” | Vendor-reported | Mar 11, 2026 | |
| 6 | MiniMax-M2.1OW | MiniMax | 88.0% | 6 | Overall accuracy; data source: Self-Reported; 229B | Vendor-reported | Mar 11, 2026 |
| 7 | Qwen3.5 397B-A17BOW | Alibaba (Qwen) | 87.8% | 7 | Overall accuracy; data source: Self-Reported; 397B“Qwen3.5-397B-A17B” | Vendor-reported | Mar 11, 2026 |
| 8 | Seed2.0-Lite | ByteDance | 87.7% | 8 | Overall accuracy; data source: Self-Reported | Vendor-reported | Mar 11, 2026 |
| 9 | GPT-5.4 | OpenAI | 87.5% | 9 | Overall accuracy; data source: Self-Reported | Vendor-reported | Mar 11, 2026 |
| 10 | GPT-5.2 | OpenAI | 87.4% | 10 | Overall accuracy; data source: Self-Reported | Vendor-reported | Mar 11, 2026 |
| 11 | Claude-4.5-Sonnet(Thinking) | Anthropic | 87.4% | 10 | Overall accuracy; data source: Self-Reported | Vendor-reported | Mar 11, 2026 |
| 12 | claude-4-opus | Anthropic | 87.3% | 12 | Overall accuracy; data source: Self-Reported“Claude-4-Opus-Thinking” | Vendor-reported | Mar 11, 2026 |
| 13 | claude-4-5-opus | Anthropic | 87.3% | 12 | Overall accuracy; data source: Self-Reported“Claude-4.5-Opus(Thinking)” | Vendor-reported | Mar 11, 2026 |
| 14 | Claude-4.6-Sonnet(Thinking) | Anthropic | 87.3% | 12 | Overall accuracy; data source: Self-Reported | Vendor-reported | Mar 11, 2026 |
| 15 | Hunyuan-T1 | Tencent | 87.2% | 15 | Overall accuracy; data source: Self-Reported; 389B | Vendor-reported | Mar 11, 2026 |
| 16 | GPT-5 | OpenAI | 87.1% | 16 | Overall accuracy; data source: Self-Reported“GPT-5(high)” | Vendor-reported | Mar 11, 2026 |
| 17 | K2.5-1T-A32B | — | 87.1% | 16 | Overall accuracy; data source: Self-Reported; 1000B | Vendor-reported | Mar 11, 2026 |
| 18 | Seed-Thinking-v1.5 | ByteDance | 87.0% | 18 | Overall accuracy; data source: Self-Reported; 200B | Vendor-reported | Mar 11, 2026 |
| 19 | Grok 4 | xAI | 87.0% | 18 | Overall accuracy; data source: Self-Reported“Grok-4” | Vendor-reported | Mar 11, 2026 |
| 20 | Seed2.0-Pro | ByteDance | 87.0% | 18 | Overall accuracy; data source: Self-Reported | Vendor-reported | Mar 11, 2026 |
| 21 | Qwen3.5-122B-A10BOW | Alibaba (Qwen) | 86.7% | 21 | Overall accuracy; data source: Self-Reported; 122B | Vendor-reported | Mar 11, 2026 |
| 22 | Seed1.6-Thinking | ByteDance | 86.6% | 22 | Overall accuracy; data source: Self-Reported; 200B | Vendor-reported | Mar 11, 2026 |
| 23 | Seed1.6-Base | ByteDance | 86.6% | 22 | Overall accuracy; data source: Self-Reported; 200B | Vendor-reported | Mar 11, 2026 |
| 24 | Seed1.6-Ada-Thinking | ByteDance | 86.4% | 24 | Overall accuracy; data source: Self-Reported; 200B | Vendor-reported | Mar 11, 2026 |
| 25 | GPT-5.1 | OpenAI | 86.4% | 24 | Overall accuracy; data source: Self-Reported | Vendor-reported | Mar 11, 2026 |
| 26 | Gemini-3.1-Flash-Lite-Preview | 86.2% | 26 | Overall accuracy; data source: Self-Reported | Vendor-reported | Mar 11, 2026 | |
| 27 | GPT-4.5 | OpenAI | 86.1% | 27 | Overall accuracy; data source: Self-Reported | Vendor-reported | Mar 11, 2026 |
| 28 | Qwen3.5-27BOW | Alibaba (Qwen) | 86.1% | 27 | Overall accuracy; data source: Self-Reported; 27B | Vendor-reported | Mar 11, 2026 |
| 29 | Gemini 2.5 Pro (Jun 2025) | 86.0% | 29 | Overall accuracy; data source: Self-Reported“Gemini-2.5-Pro” | Vendor-reported | Mar 11, 2026 | |
| 30 | GLM-5OW | Zhipu AI (Z.ai) | 86.0% | 29 | Overall accuracy; data source: Self-Reported; 754B | Vendor-reported | Mar 11, 2026 |
| 31 | Qwen3-Max | Alibaba (Qwen) | 85.7% | 31 | Overall accuracy; data source: Self-Reported“Qwen3-Max-Thinking” | Vendor-reported | Mar 11, 2026 |
| 32 | Qwen3.5-35B-A3BOW | Alibaba (Qwen) | 85.3% | 32 | Overall accuracy; data source: Self-Reported; 35B | Vendor-reported | Mar 11, 2026 |
| 33 | GPT-o3-high | OpenAI | 85.0% | 33 | Overall accuracy; data source: Self-Reported | Vendor-reported | Mar 11, 2026 |
| 34 | DeepSeek-V3.2OW | DeepSeek | 85.0% | 33 | Overall accuracy; data source: Self-Reported; 685B“DeepSeek-V3.2-Thinking” | Vendor-reported | Mar 11, 2026 |
| 35 | DeepSeek-V3.1-ThinkingOW | DeepSeek | 84.8% | 35 | Overall accuracy; data source: Self-Reported; 681B | Vendor-reported | Mar 11, 2026 |
| 36 | GLM-4.5OW | Zhipu AI (Z.ai) | 84.6% | 36 | Overall accuracy; data source: Self-Reported; 355B | Vendor-reported | Mar 11, 2026 |
| 37 | Gemini 2.5 Pro (Mar 2025) | 84.5% | 37 | Overall accuracy; data source: TIGER-Lab“Gemini-2.5-Pro-Exp-03-25” | Official leaderboard | Mar 11, 2026 | |
| 38 | Qwen3-235B-A22B-Thinking (Jul 2025)OW | Alibaba (Qwen) | 84.5% | 38 | Overall accuracy; data source: Self-Reported; 235B“Qwen3-235B-A22B-Thinking-2507” | Vendor-reported | Mar 11, 2026 |
| 39 | Grok 4.1 fast | xAI | 84.2% | 39 | Overall accuracy; data source: Self-Reported“Grok-4.1-Fast(Reasoning)” | Vendor-reported | Mar 11, 2026 |
| 40 | DeepSeek-R1OW | DeepSeek | 84.0% | 40 | Overall accuracy; data source: Self-Reported; 671B | Vendor-reported | Mar 11, 2026 |
| 41 | Claude 3.7 Sonnet | Anthropic | 84.0% | 40 | Overall accuracy; data source: Self-Reported“Claude-3.7-Sonnet-Thinking” | Vendor-reported | Mar 11, 2026 |
| 42 | claude-4-sonnet | Anthropic | 83.7% | 42 | Overall accuracy; data source: Self-Reported“Claude-4-Sonnet” | Vendor-reported | Mar 11, 2026 |
| 43 | DeepSeek-V3.1-NonThinking | DeepSeek | 83.7% | 42 | Overall accuracy; data source: Self-Reported; 671B | Vendor-reported | Mar 11, 2026 |
| 44 | Seed2.0-Mini | ByteDance | 83.6% | 44 | Overall accuracy; data source: Self-Reported | Vendor-reported | Mar 11, 2026 |
| 45 | Intern-S1 | — | 83.5% | 45 | Overall accuracy; data source: Self-Reported; 235B | Vendor-reported | Mar 11, 2026 |
| 46 | DeepSeek-R1 (May 2025)OW | DeepSeek | 83.4% | 46 | Overall accuracy; data source: Self-Reported; 671B“DeepSeek-R1-0528” | Vendor-reported | Mar 11, 2026 |
| 47 | Grok-3 mini | xAI | 83.0% | 47 | Overall accuracy; data source: Self-Reported“Grok-3-mini” | Vendor-reported | Mar 11, 2026 |
| 48 | GPT-4-mini (high) | OpenAI | 83.0% | 47 | Overall accuracy; data source: Self-Reported | Vendor-reported | Mar 11, 2026 |
| 49 | Qwen3-235B-A22B-Instruct (Jul 2025)OW | Alibaba (Qwen) | 83.0% | 47 | Overall accuracy; data source: Self-Reported; 235B“Qwen3-235B-A22B-Instruct-2507” | Vendor-reported | Mar 11, 2026 |
| 50 | Llama4-Behemoth | Meta | 82.8% | 50 | Overall accuracy; data source: Self-Reported; 2000B | Vendor-reported | Mar 11, 2026 |
| 51 | seed-oss-36b-instruct | ByteDance | 82.7% | 51 | Overall accuracy; data source: Self-Reported; 36B“Seed-OSS-36B-Instruct” | Vendor-reported | Mar 11, 2026 |
| 52 | LongCat-Flash-ChatOW | Meituan | 82.7% | 51 | Overall accuracy; data source: Self-Reported; 560B | Vendor-reported | Mar 11, 2026 |
| 53 | Qwen3.5-9BOW | Alibaba (Qwen) | 82.5% | 53 | Overall accuracy; data source: Self-Reported; 9B | Vendor-reported | Mar 11, 2026 |
| 54 | MiniMax-M2OW | MiniMax | 82.0% | 54 | Overall accuracy; data source: Self-Reported; 229B | Vendor-reported | Mar 11, 2026 |
| 55 | GPT-4.1 | OpenAI | 81.8% | 55 | Overall accuracy; data source: Self-Reported | Vendor-reported | Mar 11, 2026 |
| 56 | GLM-4.5-AirOW | Zhipu AI (Z.ai) | 81.4% | 56 | Overall accuracy; data source: Self-Reported; 106B | Vendor-reported | Mar 11, 2026 |
| 57 | DeepSeek-V3 (Mar 2025)OW | DeepSeek | 81.3% | 57 | Overall accuracy; data source: Self-Reported; 671B“Deepseek-V3-0324” | Vendor-reported | Mar 11, 2026 |
| 58 | MiniMax-M1OW | MiniMax | 81.1% | 58 | Overall accuracy; data source: Self-Reported; 456B | Vendor-reported | Mar 11, 2026 |
| 59 | Kimi K2 (Jul 2025)OW | Moonshot AI | 81.0% | 59 | Overall accuracy; data source: Self-Reported; 1000B“Kimi-K2-Instruct” | Vendor-reported | Mar 11, 2026 |
| 60 | Qwen3-30B-A3B-Thinking (Jul 2025)OW | Alibaba (Qwen) | 80.9% | 60 | Overall accuracy; data source: Self-Reported; 30B“Qwen3-30B-A3B-Thinking-2507” | Vendor-reported | Mar 11, 2026 |
| 61 | gpt-oss-120bOW | OpenAI | 80.8% | 61 | Overall accuracy; data source: Self-Reported; 120B“GPT-oss-120B(high)” | Vendor-reported | Mar 11, 2026 |
| 62 | Llama4-Maverick | Meta | 80.5% | 62 | Overall accuracy; data source: Self-Reported; 400B | Vendor-reported | Mar 11, 2026 |
| 63 | GPT-o1-mini | OpenAI | 80.3% | 63 | Overall accuracy; data source: Self-Reported | Vendor-reported | Mar 11, 2026 |
| 64 | Doubao-1.5-Pro | ByteDance | 80.1% | 64 | Overall accuracy; data source: Self-Reported | Vendor-reported | Mar 11, 2026 |
| 65 | MiniMax-M2.5OW | MiniMax | 80.1% | 64 | Overall accuracy; data source: Self-Reported; 229B | Vendor-reported | Mar 11, 2026 |
| 66 | Grok3-Beta | xAI | 79.9% | 66 | Overall accuracy; data source: Self-Reported | Vendor-reported | Mar 11, 2026 |
| 67 | GPT-o3-mini | OpenAI | 79.4% | 67 | Overall accuracy; data source: Self-Reported | Vendor-reported | Mar 11, 2026 |
| 68 | Gemini 2.0 Pro | 79.1% | 68 | Overall accuracy; data source: Self-Reported“Gemini-2.0-Pro” | Vendor-reported | Mar 11, 2026 | |
| 69 | Qwen3.5-4BOW | Alibaba (Qwen) | 79.1% | 68 | Overall accuracy; data source: Self-Reported; 4B | Vendor-reported | Mar 11, 2026 |
| 70 | HunyuanTurboS | Tencent | 79.0% | 70 | Overall accuracy; data source: Self-Reported; 560B | Vendor-reported | Mar 11, 2026 |
| 71 | Grok3-mini-Beta | xAI | 78.9% | 71 | Overall accuracy; data source: Self-Reported | Vendor-reported | Mar 11, 2026 |
| 72 | Qwen3-30B-A3BOW | Alibaba (Qwen) | 78.5% | 72 | Overall accuracy; data source: Self-Reported; 30B“Qwen3-30B-A3B-Thinking” | Vendor-reported | Mar 11, 2026 |
| 73 | ERNIE-4.5-300B-A47B | Baidu | 78.4% | 73 | Overall accuracy; data source: Self-Reported; 300B | Vendor-reported | Mar 11, 2026 |
| 74 | Nemotron-3-Nano-30B-A3B(BF16) | NVIDIA | 78.3% | 74 | Overall accuracy; data source: Self-Reported; 30B | Vendor-reported | Mar 11, 2026 |
| 75 | Nemotron-3-Nano-30B-A3B(FP8) | NVIDIA | 78.1% | 75 | Overall accuracy; data source: Self-Reported; 30B | Vendor-reported | Mar 11, 2026 |
| 76 | Claude 3.5 Sonnet | Anthropic | 77.6% | 78 | Overall accuracy; data source: TIGER-LAb“Claude-3.5-Sonnet (2024-10-22)” | Official leaderboard | Mar 11, 2026 |
| 77 | gemini-2.0-flash | 77.6% | 79 | Overall accuracy; data source: Self-Reported“Gemini-2.0-Flash” | Vendor-reported | Mar 11, 2026 | |
| 78 | Gemini 2.0 Flash (Dec 2024) | 76.2% | 80 | Overall accuracy; data source: TIGER-Lab“Gemini-2.0-Flash-exp” | Official leaderboard | Mar 11, 2026 | |
| 79 | Qwen2.5-Max | Alibaba (Qwen) | 76.1% | 82 | Overall accuracy; data source: Self-Reported | Vendor-reported | Mar 11, 2026 |
| 80 | Phi-4-reasoning-plus | Microsoft | 76.0% | 83 | Overall accuracy; data source: Self-Reported; 14B | Vendor-reported | Mar 11, 2026 |
| 81 | DeepSeek-V3OW | DeepSeek | 75.9% | 84 | Overall accuracy; data source: Self-Reported; 671B“Deepseek-V3” | Vendor-reported | Mar 11, 2026 |
| 82 | MiniMax-Text-01 | MiniMax | 75.7% | 85 | Overall accuracy; data source: Self-Reported; 456B | Vendor-reported | Mar 11, 2026 |
| 83 | Grok-2 | xAI | 75.5% | 86 | Overall accuracy; data source: Self-Reported | Vendor-reported | Mar 11, 2026 |
| 84 | Grok-4.1-Fast(Non-Reasoning) | xAI | 75.2% | 87 | Overall accuracy; data source: Self-Reported | Vendor-reported | Mar 11, 2026 |
| 85 | Llama4-Scout | Meta | 74.3% | 88 | Overall accuracy; data source: Self-Reported; 109B | Vendor-reported | Mar 11, 2026 |
| 86 | Llama 3.1-405BOW | Meta | 73.3% | 90 | Overall accuracy; data source: Self-Reported; 405B“Llama-3.1-405B-Instruct” | Vendor-reported | Mar 11, 2026 |
| 87 | gpt-oss-20bOW | OpenAI | 73.1% | 91 | Overall accuracy; data source: TIGER-Lab; 20B“GPT-oss-20B(medium)” | Official leaderboard | Mar 11, 2026 |
| 88 | Athene-V2-Chat (0-shot) | Nexusflow | 73.1% | 92 | Overall accuracy; data source: TIGER-Lab; 72B | Official leaderboard | Mar 11, 2026 |
| 89 | GPT-4o (May 2024) | OpenAI | 72.5% | 93 | Overall accuracy; data source: TIGER-Lab“GPT-4o (2024-05-13)” | Official leaderboard | Mar 11, 2026 |
| 90 | Grok-2-mini | xAI | 71.8% | 94 | Overall accuracy; data source: Self-Reported | Vendor-reported | Mar 11, 2026 |
| 91 | Gemini-2.0-Flash-Lite | 71.6% | 95 | Overall accuracy; data source: Self-Reported | Vendor-reported | Mar 11, 2026 | |
| 92 | Qwen2.5-72BOW | Alibaba (Qwen) | 71.6% | 96 | Overall accuracy; data source: Self-Reported; 72B | Vendor-reported | Mar 11, 2026 |
| 93 | ECHO_Ego_v2_14B | — | 71.2% | 97 | Overall accuracy; data source: Self-Reported; 14B | Vendor-reported | Mar 11, 2026 |
| 94 | QwQ-32B-PreviewOW | Alibaba (Qwen) | 71.0% | 98 | Overall accuracy; data source: Self-Reported; 32B | Vendor-reported | Mar 11, 2026 |
| 95 | Phi-4OW | Microsoft Research | 70.4% | 99 | Overall accuracy; data source: Self-Reported; 14B | Vendor-reported | Mar 11, 2026 |
| 96 | Gemini 1.5 Pro (Sept 2024) | 70.3% | 100 | Overall accuracy; data source: TIGER-Lab“Gemini-1.5-Pro-002” | Official leaderboard | Mar 11, 2026 | |
| 97 | Athene-V2-ChatOW | Nexusflow | 70.2% | 101 | Overall accuracy; data source: TIGER-Lab; 72B | Official leaderboard | Mar 11, 2026 |
| 98 | ERNIE-4.5-300B-A47B-Base | Baidu | 69.5% | 102 | Overall accuracy; data source: Self-Reported; 300B | Vendor-reported | Mar 11, 2026 |
| 99 | Qwen2.5-32BOW | Alibaba (Qwen) | 69.2% | 103 | Overall accuracy; data source: Self-Reported; 32B | Vendor-reported | Mar 11, 2026 |
| 100 | SkyThought-T1 | — | 69.2% | 104 | Overall accuracy; data source: Self-Reported; 32B | Vendor-reported | Mar 11, 2026 |
Data sources & licences
| Source | Licence / terms | How we reuse it | Last fetched |
|---|---|---|---|
| MMLU-Pro leaderboard (TIGER-Lab) TIGER-Lab, MMLU-Pro leaderboard | Apache-2.0 (dataset) | Open licence – full data with attribution | Oct 22, 2026 |
FAQ
What does MMLU-Pro measure?
Harder successor to MMLU: about 12,000 reasoning-heavy questions in 14 disciplines with 10 answer options. Knowledge plus reasoning, usually with chain-of-thought.
How reliable is MMLU-Pro?
The official leaderboard mixes TIGER-Lab runs with self-reported numbers; we label each row. Contamination risk: Medium – Public test set since 2024.