Qwen 2.5 Coder 14B Instruct: benchmark results
Alibaba's Apache-2.0 14B code model from the November 2024 Qwen2.5-Coder expansion to six sizes, trained on 5.5T code-heavy tokens with 128K context. Provider: Alibaba. Released 2024-11-12. Access: Open.
Aggregate Rating 1532 ± 12, rank #675 of 1656 rated models, from 91 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| GSM8K | 94.2 | Accuracy (%) | 98.9 |
| Open Korean LLM - Ko-GSM8K (flexible extract) | 72.18 | Exact match, flexible extract (%) | 98.6 |
| Open Korean LLM - Ko-IFEval | 64 | Strict accuracy, prompt/instruction mean (%) | 98 |
| Open Korean LLM Leaderboard | 49.63 | Average Score (%) | 97 |
| Open Ko-LLM Leaderboard | 49.63 | Average (%) | 96.9 |
| Open Japanese LLM - Wiki PAS SET F1 | 11.46 | Score (%) | 94.7 |
| Open Portuguese LLM - ASSIN2 RTE | 94.07 | Macro F1 (%) | 94.4 |
| Open Korean LLM - Ko-EQ-Bench | 44.34 | EQ-Bench score | 93.5 |
| BigCodeBench | 48.2 | Pass@1 (%) | 92 |
| Delulu - CodeBLEU | 51.5 | CodeBLEU (0-100): n-gram, syntax-tree and dataflow agreement | 90 |
| Open Portuguese LLM - BLUEX | 66.48 | Accuracy (%) | 88.5 |
| Open Korean LLM - KorNAT-Harmless | 67.14 | Normalized accuracy (%) | 88.1 |
Interactive version: theaggregate.ai/model?slug=qwen-2-5-coder-14b-instruct · How It Works · Data refreshed daily, snapshot 2026-10-20.