Artificial Analysis's composite score for coding ability, combining multiple code-related benchmarks.
Aggregates multiple coding benchmarks for a quick top-line comparison across models; smooths noise from any individual coding leaderboard update.
Composite score hides whether gains come from contest coding, bug fixing, or agentic terminal work; weighting can overrepresent benchmark families that do not match your production tasks.
Higher is better. Composite coding score — use for relative ranking between models.
Composite score hides whether gains come from contest coding, bug fixing, or agentic terminal work.
| # | Model | Score |
|---|---|---|
| 1 | 76.3 | |
| 2 | 76.2 | |
| 3 | 68.9 | |
| 4 | Agnes 2.5 Pro Alpha (based on Qwen3.5 397B A17B) source artificial_analysis Evidence observed Sep 24, 2026 | 58.8 |
| 5 | 53.1 | |
| 6 | 52.7 | |
| 7 | 51.5 | |
| 8 | 45.6 | |
| 9 | 44.9 | |
| 10 | 43.9 | |
| 11 | 42.1 | |
| 12 | 41.9 | |
| 13 | 41.3 | |
| 14 | 37.8 | |
| 15 | 36.8 | |
| 16 | 36.5 | |
| 17 | 36.2 | |
| 18 | 35.5 | |
| 19 | 33.5 | |
| 20 | 31.7 | |
| 21 | 30.6 | |
| 22 | 29.6 | |
| 23 | 26.3 | |
| 24 | 21.4 | |
| 25 | 16.4 | |
| 26 | 14.2 | |
| 27 | 10.9 | |
| 28 | 10.8 | |
| 29 | 10.7 | |
| 30 | 10.7 | |
| 31 | 10.7 | |
| 32 | 7.8 | |
| 33 | 6.8 | |
| 34 | 5.6 | |
| 35 | 4.9 | |
| 36 | 4.3 | |
| 37 | 4.0 | |
| 38 | 0.6 |