DeepSeek Reasoner: benchmark results
DeepSeek's deepseek-reasoner API endpoint row, introduced with the R1 reasoning model in January 2025. Provider: DeepSeek. Released 2025-01-20. Access: API.
Aggregate Rating 1684 ± 16, rank #236 of 1640 rated models, from 56 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Nejumi 4 - ALT - Bias | 95.46 | Score (%) | 94 |
| Nejumi 4 - ALT - Toxicity | 85.09 | Score (%) | 87.2 |
| Nejumi 4 - ALT Average | 88.66 | Score (%) | 84.2 |
| BIRD-CRITIC | 33.67 | Score | 80 |
| Nejumi 4 - GLP - Expression | 98 | Score (%) | 77.8 |
| Nejumi 4 - ALT - Truthfulness | 80.97 | Score (%) | 76.7 |
| IslamicTurathBench | 77.63 | IslamicTurathBench Score (%) | 75 |
| OTIS Mock AIME 2024-25 | 87.82 | Accuracy (%) | 73 |
| Xent Games | 62.67 | Overall Score (self-reported) | 72.7 |
| Mizan LLM Leaderboard | 65.49 | Average Score (0-100) | 70.5 |
| Nejumi 4 - GLP - Mathematical Reasoning | 95.5 | Score (%) | 70.3 |
| Epoch AI - GPQA Diamond | 83.42 | Accuracy (%) | 67.5 |
Interactive version: theaggregate.ai/model?slug=deepseek-reasoner · How It Works · Data refreshed daily, snapshot 2026-10-13.