MMLU-Pro leaderboard
5 models have a published score. DeepSeek V4 Pro (DeepSeek) leads with 87.5, and the spread to the lowest score, Gemma 4 26B A4B at 82.6, is 4.9 points. The highest-scoring model you can deploy on OpenLLM Buddy is Qwen3.6 27B, ranked 3 of 5 with 86.2.
Best closed api
No published score
0 models scored
MMLU-Pro scores
Accuracy. With ten options, random guessing scores 10%.
- OpenLLM Buddy catalog
- Open weights
- Closed API
What it measures
More than 12,000 questions across 14 domains, from law and economics to engineering and health, rebuilt from MMLU with harder distractors and more reasoning-heavy items.
How it is scored
Accuracy. With ten options, random guessing scores 10%.
Why it matters
It is a stable, widely reported read on general knowledge after the original MMLU saturated.
Official benchmark pageMore knowledge tests
Sources
| Model | Test | Score | Setting | Source |
|---|---|---|---|---|
| DeepSeek V4 Pro | MMLU-Pro | 87.5 | V4-Pro (Preview) Max, EM | Hugging Face model card (DeepSeek-V4-Pro) |
| Nemotron 3 Ultra | MMLU-Pro | 86.8 | — | Hugging Face model card (NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16) |
| Qwen3.6 27B | MMLU-Pro | 86.2 | — | Qwen3.6-27B model card (Hugging Face) |
| Gemma 4 31B | MMLU-Pro | 85.2 | — | Gemma 4 31B IT model card (Hugging Face) |
| Gemma 4 26B A4B | MMLU-Pro | 82.6 | — | Gemma 4 31B IT model card (Hugging Face), 26B A4B column |
Scores are published results from model cards, lab reports, and public leaderboards, each linked to its source. Settings such as thinking mode or test version differ between sources and are shown next to each score. OpenLLM Buddy catalog models are uncensored fine-tunes; their scores are those of the official base model. Data collected 2026-10-30. How the index works