DeepSeek R1 Distill Qwen 14B: benchmark results
DeepSeek's official R1 reasoning distillation onto Qwen2.5-14B, MIT-licensed and strong on math and code. Provider: DeepSeek. Released 2025-01-20. Access: Open.
Aggregate Rating 1481 ± 8, rank #844 of 1656 rated models, from 536 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| FACTS Leaderboard | 45.76 | Combined Score (%) | 100 |
| Open CoT - LogiQA | 13.74 | CoT Gain (%) | 100 |
| Open CoT Leaderboard | 17.65 | Average CoT Gain (%) | 100 |
| Open LLM Leaderboard - MuSR | 53.66 | Score | 99.8 |
| Open LLM Leaderboard - MATH Level 5 | 57.02 | Score | 99.5 |
| Open CoT - LSAT Analytical Reasoning | 16.96 | CoT Gain (%) | 99.2 |
| Open CoT - LogiQA 2 | 18.7 | CoT Gain (%) | 99.2 |
| Open Portuguese LLM - ENEM | 84.25 | Accuracy (%) | 97.1 |
| Open LLM Leaderboard - GPQA | 38.76 | Score | 96.9 |
| Open Portuguese LLM - BLUEX | 76.5 | Accuracy (%) | 96.3 |
| RobustBench-TC - Misleading Description | 42.2 | Tool-call accuracy (%) under the MisDesc perturbation (rewar | 94.1 |
| RobustBench-TC - Rate Limit | 37.7 | Tool-call accuracy (%) under the RateLim perturbation (trans | 94.1 |
Interactive version: theaggregate.ai/model?slug=deepseek-r1-distill-qwen-14b · How It Works · Data refreshed daily, snapshot 2026-10-20.