MMLU-Redux
We show this table for reference; we do not rank on it.
A harder refresh of MMLU intended to keep broad knowledge evaluation useful after the original benchmark became too easy for frontier models.
Benchmark score on MMLU-Redux — November 13, 2026
We compile the MMLU-Redux rows from provider self-reports and secondary reports. Claude Opus 4.5 leads the table at 96.6%, followed by Qwen3.7 Max (95%) and Qwen3.5 397B (94.9%). We do not use these results to rank models overall.
Claude Opus 4.5
Anthropic
Another lab’s comparison tableQwen3.7 Max
Alibaba
Reported by AlibabaQwen3.5 397B
Alibaba
Reported by Alibaba21 modelsKnowledgeCurrentDisplay onlyUpdated November 13, 2026
Benchmark score table (21 models)
None of the 21 results on this page has a documented run by the benchmark or an independent evaluator: 19 reported by the labs that built the models and 2 reported in another publisher’s document.
| Rank | Model / configuration | Score | Who ran it | Parameters (B) | Open / closed |
|---|---|---|---|---|---|
| 1 | Claude Opus 4.5AnthropicAnother lab’s comparison table | 96.6% | Another lab’s comparison table | Not reported | Closed |
| 2 | Qwen3.7 MaxAlibabaReported by Alibaba | 95% | Reported by Alibaba | Not reported | Closed |
| 3 | Qwen3.5 397BAlibabaReported by Alibaba | 94.9% | Reported by Alibaba | Not reported | Open |
| 4 | Qwen3.7 PlusAlibabaReported by Alibaba | 94.5% | Reported by Alibaba | Not reported | Closed |
| 5 | Qwen3.6 PlusAlibabaReported by Alibaba | 94.5% | Reported by Alibaba | Not reported | Closed |
| 6 | Kimi K2.5Moonshot AIAnother lab’s comparison table | 94.5% | Another lab’s comparison table | Not reported | Open |
| 7 | Qwen3.6-27BAlibabaReported by Alibaba | 93.5% | Reported by Alibaba | Not reported | Open |
| 8 | DeepSeek-R1-0528DeepSeekReported by DeepSeek | 93.4% | Reported by DeepSeek | Not reported | Open |
| 9 | Qwen3 235B-A22B (Reasoning)AlibabaReported by Alibaba | 92.7% | Reported by Alibaba | Not reported | Open |
| 10 | Qwen3 32B (Reasoning)AlibabaReported by Alibaba | 90.9% | Reported by Alibaba | Not reported | Open |
| 11 | DeepSeek V4 Pro BaseDeepSeekReported by DeepSeek | 90.8% | Reported by DeepSeek | Not reported | Open |
| 12 | QwQ-32BAlibabaReported by Alibaba | 90% | Reported by Alibaba | Not reported | Open |
| 13 | DeepSeek V4 Flash BaseDeepSeekReported by DeepSeek | 89.4% | Reported by DeepSeek | Not reported | Open |
| 14 | Qwen3 235B-A22BAlibabaReported by Alibaba | 89.2% | Reported by Alibaba | Not reported | Open |
| 15 | Ternary Bonsai 2 27BPrism MLReported by Prism ML | 89.1% | Reported by Prism ML | Not reported | Open |
| 16 | Mellum2.1-12B-A2.5B-ThinkingJetBrainsReported by JetBrains | 87.8% | Reported by JetBrains | Not reported | Open |
| 17 | Mellum2-12B-A2.5B-ThinkingJetBrainsReported by JetBrains | 86.2% | Reported by JetBrains | Not reported | Open |
| 18 | Qwen3 32BAlibabaReported by Alibaba | 85.7% | Reported by Alibaba | Not reported | Open |
| 19 | MiniCPM5-2BOpenBMBReported by OpenBMB | 84.7% | Reported by OpenBMB | Not reported | Open |
| 20 | Mellum2-12B-A2.5B-InstructJetBrainsReported by JetBrains | 78.1% | Reported by JetBrains | Not reported | Open |
| 21 | MiniCPM5-1BOpenBMBReported by OpenBMB | 70.1% | Reported by OpenBMB | Not reported | Open |
- Lab-reported
- Benchmark-run
- Independent
Among the reported MMLU-Redux rows, Claude Opus 4.5 is first at 96.6%. The third row is 1.7 points behind. The broader top-10 range is 5.7 points, so many of the published results sit in a relatively narrow band.
21 models have been evaluated on MMLU-Redux. The benchmark falls in the Knowledge category. MMLU-Redux is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.
About MMLU-Redux
Year
2026
Tasks
Broad academic QA
Format
Multiple choice questions
Difficulty
Advanced general knowledge
MMLU-Redux is useful when MMLU itself has largely saturated. It acts as a broader knowledge sanity check with fresher or harder questions intended to preserve separation among strong general-purpose models.
Freshness and provenance
Version
MMLU-Redux 2026
Refresh cadence
Quarterly
Staleness state
Current
Question availability
Public benchmark set
BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.
Questions
What does MMLU-Redux measure?
A harder refresh of MMLU intended to keep broad knowledge evaluation useful after the original benchmark became too easy for frontier models.
Which model scores highest on MMLU-Redux?
Claude Opus 4.5 by Anthropic currently leads with a score of 96.6% on MMLU-Redux.
How many models are evaluated on MMLU-Redux?
21 AI models have published results on MMLU-Redux in the BenchLM catalog.
Compare top models on MMLU-Redux
Know when it’s worth switching models
The model to choose, the cheaper alternative, and the release we would wait on.
Read a sample issueJoin 7,000+ subscribers.
One email each week. Unsubscribe anytime.