LLMs TigerLive LLM benchmarks

MMLU-Pro leaderboard: which LLMs score highest

MMLU-Pro (57-subject knowledge + reasoning) currently has scores for 136 models. The top score is 88.1% by Darwin-180B-RSI; the median model scores 55.3% and the lowest scores 10.8%.

MMLU-Pro leaderboard: top 50 of 136 models (highest-effort row per model)
#ModelOrganisationMMLU-Pro scoreCheapest input $/MSourceCaptured
1Darwin-180B-RSIFINAL-Bench88.1%—HuggingFace MMLU-Pro leaderboard2026-10-05
2Intern-S2-Previewinternlm88.0%—HuggingFace MMLU-Pro leaderboard2026-10-05
3MiniMax-M2.1MiniMax88.0%$0.300HuggingFace MMLU-Pro leaderboard2026-10-05
4Qwen3.5 397B-A17BAlibaba87.8%$0.164HuggingFace MMLU-Pro leaderboard2026-10-05
5Kimi K2.5Moonshot87.1%$0.450HuggingFace MMLU-Pro leaderboard2026-10-05
6NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16nvidia86.8%—HuggingFace MMLU-Pro leaderboard2026-10-05
7NVIDIA-Nemotron-3-Ultra-550B-A55B-NVFP4nvidia86.8%—HuggingFace MMLU-Pro leaderboard2026-10-05
8Qwen3.5-122B-A10BQwen86.7%$0.113HuggingFace MMLU-Pro leaderboard2026-10-05
9Qwen3.6 27BAlibaba86.2%$0.150HuggingFace MMLU-Pro leaderboard2026-10-05
10Solar-Open2-250Bupstage86.2%—HuggingFace MMLU-Pro leaderboard2026-10-05
11Qwen3.5 27BAlibaba86.1%$0.195HuggingFace MMLU-Pro leaderboard2026-10-05
12GLM-5Z.ai (Zhipu AI)86.0%$0.600HuggingFace MMLU-Pro leaderboard2026-10-05
13Qwen3.6 35B-A3BAlibaba85.2%$0.050HuggingFace MMLU-Pro leaderboard2026-10-05
14DeepSeek R1 (0528)deepseek-ai85.0%$0.400HuggingFace MMLU-Pro leaderboard2026-10-05
15glm-4.5Z.ai (Zhipu AI),Tsinghua University84.6%$0.400HuggingFace MMLU-Pro leaderboard2026-10-05
16Step-3.5-Flashstepfun-ai84.4%$0.100HuggingFace MMLU-Pro leaderboard2026-10-05
17DeepSeek-R1DeepSeek84.0%$0.400HuggingFace MMLU-Pro leaderboard2026-10-05
18K-EXAONE-236B-A23BLGAI-EXAONE83.8%—HuggingFace MMLU-Pro leaderboard2026-10-05
19NVIDIA-Nemotron-3-Super-120B-A12B-BF16nvidia83.7%—HuggingFace MMLU-Pro leaderboard2026-10-05
20Intern-S1internlm83.5%—HuggingFace MMLU-Pro leaderboard2026-10-05
21EXAONE-4.5-33BLGAI-EXAONE83.3%—HuggingFace MMLU-Pro leaderboard2026-10-05
22Qwen3-235B-A22B-Instruct-2507Alibaba83.0%$0.087HuggingFace MMLU-Pro leaderboard2026-10-05
23LongCat-Flash-Chatmeituan-longcat82.7%—HuggingFace MMLU-Pro leaderboard2026-10-05
24seed-oss-36b-instructByteDance-Seed82.7%—HuggingFace MMLU-Pro leaderboard2026-10-05
25MiniMax-M2MiniMax82.0%$0.300HuggingFace MMLU-Pro leaderboard2026-10-05
26NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16nvidia81.9%—HuggingFace MMLU-Pro leaderboard2026-10-05
27NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4nvidia81.6%—HuggingFace MMLU-Pro leaderboard2026-10-05
28A.X-K1skt81.5%—HuggingFace MMLU-Pro leaderboard2026-10-05
29GLM-4.5-AirZ.ai (Zhipu AI),Tsinghua University81.4%$0.125HuggingFace MMLU-Pro leaderboard2026-10-05
30DeepSeek V3 0324DeepSeek81.2%$0.200HuggingFace MMLU-Pro leaderboard2026-10-05
31MiniMax-M1-40kMiniMaxAI81.1%—HuggingFace MMLU-Pro leaderboard2026-10-05
32JoyAI-LLM-Flashjdopensource81.0%—HuggingFace MMLU-Pro leaderboard2026-10-05
33Kimi K2 InstructMoonshot81.0%$0.500HuggingFace MMLU-Pro leaderboard2026-10-05
34qwen3-30b-a3b-thinking-2507Alibaba80.9%$0.200HuggingFace MMLU-Pro leaderboard2026-10-05
35MiniMax-M2.5MiniMax80.1%$0.270HuggingFace MMLU-Pro leaderboard2026-10-05
36ERNIE-4.5-300B-A47B-PTbaidu78.4%—HuggingFace MMLU-Pro leaderboard2026-10-05
37LongCat-Flash-Litemeituan-longcat78.3%—HuggingFace MMLU-Pro leaderboard2026-10-05
38MiniMax-Text-01MiniMaxAI75.7%—HuggingFace MMLU-Pro leaderboard2026-10-05
39Qwen2.5-72BAlibaba71.6%—HuggingFace MMLU-Pro leaderboard2026-10-05
40phi-4Microsoft Research70.4%$0.070HuggingFace MMLU-Pro leaderboard2026-10-05
41qwen3-4b-instruct-2507Qwen69.6%—HuggingFace MMLU-Pro leaderboard2026-10-05
42ERNIE-4.5-300B-A47B-Base-PTbaidu69.5%—HuggingFace MMLU-Pro leaderboard2026-10-05
43Qwen2.5-32BAlibaba69.2%—HuggingFace MMLU-Pro leaderboard2026-10-05
44mimo-v2.5-proXiaomi Corp68.5%$0.435HuggingFace MMLU-Pro leaderboard2026-10-05
45qwen3-235b-a22bAlibaba68.2%$0.180HuggingFace MMLU-Pro leaderboard2026-10-05
46Mistral-Large-Instruct-2411mistralai67.9%—HuggingFace MMLU-Pro leaderboard2026-10-05
47Hunyuan-A13B-Instructtencent67.3%$0.140HuggingFace MMLU-Pro leaderboard2026-10-05
48Mistral-Large-Instruct-2407mistralai65.9%—HuggingFace MMLU-Pro leaderboard2026-10-05
49DeepSeek-V2.5deepseek-ai65.8%—HuggingFace MMLU-Pro leaderboard2026-10-05
50NVIDIA-Nemotron-3-Nano-30B-A3B-Base-BF16nvidia65.1%—HuggingFace MMLU-Pro leaderboard2026-10-05

Compare all benchmarks side by side on the live leaderboard →

Among the ten highest scorers, the cheapest listed API price belongs to Qwen3.5-122B-A10B at $0.113 per million input tokens (score 86.7%).

What MMLU-Pro measures

MMLU-Pro extends MMLU with harder, more reasoning-heavy questions and ten answer options instead of four. That lowers the chance of guessing and makes small score gaps more meaningful than on the original MMLU.

How to read these scores

Scores are percentages (higher is better): the share of questions or tasks the model completed correctly. Where a model is listed at several reasoning efforts, only its highest-effort row is shown. See the methodology for how rows are chosen.

Where the data comes from

HuggingFace MMLU-Pro leaderboard. Scores were last captured 2026-10-05; the Captured column gives each row's date.

Related benchmarks