MMLU-Pro
- DeepSeek
Results
| Model | Score | Source | Human-verified |
|---|---|---|---|
| DeepSeek-V4-Pro24 Apr 2026 · DeepSeek | 87.5% | DeepSeek V4 Preview ReleaseMax reasoning effortSource image | Yes |
Benchmark
huggingface.coHigher is better
TIGER-Lab, University of WaterlooIntroduced 3 Jun 2024