RACE — Reading comprehension benchmark (ReAding Comprehension dataset from Examinations).
| # | Model | Vendor | Best score | Runs | Last seen |
|---|---|---|---|---|---|
| 1 | Mistral Large 3 675B Base 2512 | Mistral | 93.3 | 1 | 2026-06-05 |
| 2 | DeepSeek-V3.2-Exp-Base | DeepSeek | 93.2 | 1 | 2026-06-05 |
| 3 | GLM-4.5-Base | Z.ai | 92.2 | 1 | 2026-06-05 |
| 4 | Nemotron 3 Ultra 550B A55B Base | NVIDIA | 92.2 | 1 | 2026-06-05 |
| 5 | NVIDIA-Nemotron-3-Ultra-550B-A55B-Base-BF16 | NVIDIA | 92.2 | 1 | 2026-08-24 |
| 6 | Kimi K2 Base | Moonshot | 92 | 1 | 2026-06-05 |
Best tracked score per model (default configuration; source-attributed and verification-tiered). Open a model for its full benchmark surface, provenance and pricing.