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.

Unified ELO 1432 ± 13, rank #1079 of 1776 rated models, from 347 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
FACTS Leaderboard45.76Combined Score (%)100
Open CoT - LogiQA13.74CoT Gain (%)100
Open CoT Leaderboard17.65Average CoT Gain (%)100
Open LLM Leaderboard - MuSR28.71Score99.8
Open LLM Leaderboard - MATH Level 557.02Score99.5
Open CoT - LSAT Analytical Reasoning16.96CoT Gain (%)99.2
Open CoT - LogiQA 218.7CoT Gain (%)99.2
Open LLM Leaderboard - GPQA18.34Score96.8
Open CoT - LSAT Reading Comprehension21.19CoT Gain (%)90.1
Open Japanese LLM - EL58.24Score (%)86.8
Open LLM Leaderboard - MMLU-Pro40.74Score85.7
Open CoT - LSAT Logical Reasoning17.65CoT Gain (%)82.4

Interactive version: theaggregate.ai/model?slug=deepseek-r1-distill-qwen-14b · How the rankings work · Data refreshed daily, snapshot 2026-07-22.