MMMLU
We show this table for reference; we do not rank on it.
A multilingual MMLU-style benchmark reported in provider evaluation tables.
Benchmark score on MMMLU — November 13, 2026
We compile the MMMLU rows from provider self-reports. Interfaze Beta leads the table at 90.9%, followed by Qwen3.7 Max (90.3%) and DeepSeek V4 Pro Base (90.3%). We do not use these results to rank models overall.
Interfaze Beta
Interfaze
Reported by InterfazeQwen3.7 Max
Alibaba
Reported by AlibabaDeepSeek V4 Pro Base
DeepSeek
Reported by DeepSeek17 modelsKnowledgeCurrentDisplay onlyUpdated November 13, 2026
Benchmark score table (17 models)
All 17 results on this page were reported by the labs that built the models.
| Rank | Model / configuration | Score | Who ran it | Parameters (B) | Open / closed |
|---|---|---|---|---|---|
| 1 | Interfaze BetaInterfazeReported by Interfaze | 90.9% | Reported by Interfaze | Not reported | Closed |
| 2 | Qwen3.7 MaxAlibabaReported by Alibaba | 90.3% | Reported by Alibaba | Not reported | Closed |
| 3 | DeepSeek V4 Pro BaseDeepSeekReported by DeepSeek | 90.3% | Reported by DeepSeek | Not reported | Open |
| 4 | Qwen3.7 PlusAlibabaReported by Alibaba | 89.0% | Reported by Alibaba | Not reported | Closed |
| 5 | Claude 4 Opus ThinkingAnthropicReported by Anthropic | 88.8% | Reported by Anthropic | Not reported | Closed |
| 6 | DeepSeek V4 Flash BaseDeepSeekReported by DeepSeek | 88.8% | Reported by DeepSeek | Not reported | Open |
| 7 | Claude 4 OpusAnthropicReported by Anthropic | 87.4% | Reported by Anthropic | Not reported | Closed |
| 8 | K-EXAONE 2.0LG AI ResearchReported by LG AI Research | 86.6% | Reported by LG AI Research | Not reported | Open |
| 9 | Claude 3.7 Sonnet ThinkingAnthropicReported by Anthropic | 86.1% | Reported by Anthropic | Not reported | Closed |
| 10 | Qwen3 235B-A22B (Reasoning)AlibabaReported by Alibaba | 84.3% | Reported by Alibaba | Not reported | Open |
| 11 | Gemma 4 12BGoogleReported by Google | 83.4% | Reported by Google | Not reported | Open |
| 12 | Claude 3.7 SonnetAnthropicReported by Anthropic | 83.2% | Reported by Anthropic | Not reported | Closed |
| 13 | QwQ-32BAlibabaReported by Alibaba | 80.9% | Reported by Alibaba | Not reported | Open |
| 14 | Qwen3 32B (Reasoning)AlibabaReported by Alibaba | 80.6% | Reported by Alibaba | Not reported | Open |
| 15 | Qwen3 235B-A22BAlibabaReported by Alibaba | 79.8% | Reported by Alibaba | Not reported | Open |
| 16 | Qwen3 32BAlibabaReported by Alibaba | 76.5% | Reported by Alibaba | Not reported | Open |
| 17 | Phi-3.5-miniMicrosoftReported by Microsoft | 55.4% | Reported by Microsoft | Not reported | Open |
- Lab-reported
- Benchmark-run
- Independent
Among the reported MMMLU rows, Interfaze Beta is first at 90.9%. The third row is 0.6 points behind. The broader top-10 range is 6.6 points, so many of the published results sit in a relatively narrow band.
17 models have been evaluated on MMMLU. The benchmark falls in the Knowledge category. MMMLU is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.
About MMMLU
Year
2026
Tasks
Multilingual academic QA
Format
Exact match
Difficulty
Broad multilingual knowledge
BenchLM stores MMMLU as a display-only provider-table row when exact public values are published.
Freshness and provenance
Version
MMMLU 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 MMMLU measure?
A multilingual MMLU-style benchmark reported in provider evaluation tables.
Which model scores highest on MMMLU?
Interfaze Beta by Interfaze currently leads with a score of 90.9% on MMMLU.
How many models are evaluated on MMMLU?
17 AI models have published results on MMMLU in the BenchLM catalog.
Compare top models on MMMLU
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.