Inference

MMLU-Redux

A harder refresh of MMLU intended to keep broad knowledge evaluation useful after the original benchmark became too easy for frontier models.

  • In index
  • Knowledge
ModelsModels with a result.11
Top resultBest result on this test.96.6%Claude Opus 4.5
Top-3 spreadPoints from first to third.1.7 pts
YearYear of release.N/A

Result and price

020406080100$0.3$1$3$10$30RESULTOUTPUT PRICE PER 1M TOKENS · LOG SCALE

Results

11 results
#ModelResultThe score from the source.
01Claude Opus 4.596.6%
02Qwen3.7 Max95.0%
03Qwen3.5 397B94.9%
04Qwen3.7 Plus94.5%
04Qwen3.6 Plus94.5%
06Qwen3.6-27B93.5%
07Ternary Bonsai 2 27B89.1%
08Mellum2-12B-A2.5B-Thinking86.2%
09MiniCPM5-2B84.7%
10Mellum2-12B-A2.5B-Instruct78.1%
11MiniCPM5-1B70.1%