benchgap
Multilingual

MMLU-ProX leaderboard

As of 2026-11-21, the highest measured score on MMLU-ProX is 87.0% by Qwen3.7 Max. 38 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#ModelScoreSource
1Ling 3.1 Flash90.7%estimated ± 1.3 pp, medium confidence
2Claude Opus 4.888.7%estimated ± 1.3 pp, medium confidence
3Hy4 preview88.0%estimated ± 1.3 pp, medium confidence
4Inkling 2 Small87.2%estimated ± 1.3 pp, medium confidence
5Qwen3.8-Flash-Next87.1%estimated ± 1.3 pp, medium confidence
6Qwen3.7 Max87.0%measured
7Qwen3.8-Omni-Flash86.9%estimated ± 1.3 pp, medium confidence
8Composer 2.586.6%estimated ± 1.3 pp, medium confidence
9Ornith-1.5-397B86.6%estimated ± 1.3 pp, medium confidence
10Ornith-1.0-397B86.3%estimated ± 1.3 pp, medium confidence
11Laguna S 2.186.1%estimated ± 1.3 pp, medium confidence
12Claude Sonnet 586.0%estimated ± 1.3 pp, high confidence
13Beam 1.0 501B-A23B86.0%estimated ± 1.3 pp, high confidence
14Beam 1.0 501B-A23B FP886.0%estimated ± 1.3 pp, high confidence
15Grok 4.585.9%estimated ± 1.3 pp, high confidence
16SWE-1.785.8%estimated ± 1.3 pp, high confidence
17Beam 1.0 501B-A23B NVFP485.7%estimated ± 1.3 pp, high confidence
18Claude Opus 4.585.7%measured
19Kimi K2.685.4%estimated ± 1.3 pp, high confidence
20Qwen3.7 Plus85.4%measured
21MiniMax M2.785.4%estimated ± 1.3 pp, high confidence
22DeepSeek V4 Pro85.3%estimated ± 1.3 pp, high confidence
23dots3-note Preview85.1%estimated ± 1.3 pp, high confidence
24Qwen3.5 397B84.7%measured
25Qwen3.6 Plus84.7%measured
26K-EXAONE 2.084.5%estimated ± 1.0 pp, medium confidence
27Composer 284.4%estimated ± 1.3 pp, high confidence
28Inkling 2 Mini84.4%estimated ± 1.3 pp, high confidence
29DeepSeek V4 Flash 073184.3%estimated ± 1.3 pp, high confidence
30Sage-A1.584.1%estimated ± 1.3 pp, high confidence
31Ling 3.0 Flash84.0%estimated ± 1.3 pp, high confidence
32MiMo-V2-Flash83.8%estimated ± 1.3 pp, high confidence
33Ornith-1.5-35B-A3B83.8%estimated ± 1.3 pp, high confidence
34Qwen3.6-27B83.7%estimated ± 1.3 pp, high confidence
35Ornith-1.0-35B83.2%estimated ± 1.3 pp, high confidence
36GLM-583.1%measured
37Nemotron 3 Ultra83.0%measured
38Nemotron 3.5 Super VL 120B A12B82.8%estimated ± 1.3 pp, high confidence
39Qwen3.6-35B-A3B82.7%estimated ± 1.3 pp, medium confidence
40Nemotron 3.5 Super VL 120B A12B NVFP482.7%estimated ± 1.3 pp, medium confidence
41Kimi K2.582.3%measured
42Spark-X2.6-4B82.2%estimated ± 1.3 pp, medium confidence
43Qwen3.5-122B-A10B82.2%measured
44Qwen3.5-27B82.2%measured
45Laguna M.181.9%estimated ± 1.3 pp, medium confidence
46Laguna XS 2.181.9%estimated ± 1.3 pp, medium confidence
47LongCat-Flash-Lite-Sparse81.4%estimated ± 1.3 pp, medium confidence
48Laguna XS.281.2%estimated ± 1.3 pp, medium confidence
49Qwen3.5-35B-A3B81.0%measured
50Qwen3 235B 250779.4%measured
51GPT-4.172.7%measured
52DeepSeek V3 032470.5%measured
53GPT-4o61.1%measured
54Phi-449.9%measured

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General