math benchmark
MMLU-Redux
An improved version of the MMLU benchmark featuring manually re-annotated questions to identify and correct errors in the original dataset. Provides more reliable evaluation metrics for language models by addressing dataset quality issues found in the original MMLU.
Updated Aug 7, 2026
Published models48
Registry coverage48
MetricScore
EvidenceB
MMLU-Redux leaderboard
Sorted by the source-provided rank. Higher score is better according to the registry.
Score distribution
Top published rows on the benchmark's original scale.
MMLU-Redux highlights
The top published results on this benchmark's own scale.
What is MMLU-Redux?
Definition and scoring fields from the benchmark registry.
An improved version of the MMLU benchmark featuring manually re-annotated questions to identify and correct errors in the original dataset. Provides more reliable evaluation metrics for language models by addressing dataset quality issues found in the original MMLU.
Scores are shown in ratio. The current registry marks this benchmark as not independently verified with evidence level B.
- Family
- MMLU-Redux
- Modality
- text
- Primary category
- math
- Score direction
- higher
- LLMBoard eligible
- Yes
- Evaluation key
- mmlu-redux|llm-stats-current
Source-native results are preserved. Eligibility for the overall LLMBoard score is a separate policy decision.
FAQ
Common questions about MMLU-Redux.
Which model scores highest on MMLU-Redux?
Qwen3.7 Max is currently ranked first with 95.0%.
What does MMLU-Redux measure?
An improved version of the MMLU benchmark featuring manually re-annotated questions to identify and correct errors in the original dataset. Provides more reliable evaluation metrics for language models by addressing dataset quality issues found in the original MMLU.
Is a higher score better?
Yes. Higher values rank better for this benchmark.
How many models are compared?
48 unique published model results are currently shown.
Does this benchmark affect the overall score?
This benchmark is marked as eligible for the current LLMBoard capability methodology.