Measures LLM coding ability across multiple programming languages using the Aider AI coding assistant's polyglot benchmark suite. Tests code generation, editing, and debugging in a realistic IDE-like workflow.
This definition has not completed editorial and provenance review. Verify its metric, version, source, and evaluation configuration upstream before using it for a model-selection decision.
Multi-language coverage (Python, JS, C++, Java, etc.). Tests realistic coding-assistant interactions, not just isolated functions.
Scores are 0–1 ratio. Tightly coupled to Aider's prompting strategy — scores may not generalize to other coding frameworks.
Higher is better. Pass rate ratio (0–1). Top models: ~0.83.
Scores are 0–1 ratio.
| # | Model | Score |
|---|---|---|
| 1 | 88.0 | |
| 2 | 86.7 | |
| 3 | 84.9 | |
| 4 | 81.3 | |
| 5 | 79.6 | |
| 6 | 76.9 | |
| 7 | 72.9 | |
| 8 | 72.0 | |
| 9 | 72.0 | |
| 10 | 71.4 | |
| 11 | 70.2 | |
| 12 | 64.9 | |
| 13 | 61.7 | |
| 14 | 61.3 | |
| 15 | 60.4 | |
| 16 | 59.6 | |
| 17 | 59.6 | |
| 18 | 59.1 | |
| 19 | 59.1 | |
| 20 | 56.9 | |
| 21 | 55.1 | |
| 22 | 53.3 | |
| 23 | 52.4 | |
| 24 | 51.6 | |
| 25 | 49.3 | |
| 26 | 48.4 | |
| 27 | 44.9 | |
| 28 | 41.8 | |
| 29 | 40.0 | |
| 30 | 32.9 | |
| 31 | 32.4 | |
| 32 | 28.0 | |
| 33 | 27.1 | |
| 34 | 23.1 | |
| 35 | 22.2 | |
| 36 | 21.8 | |
| 37 | 20.9 | |
| 38 | 17.8 | |
| 39 | 16.5 | |
| 40 | 16.4 | |
| 41 | 15.6 | |
| 42 | 12.0 | |
| 43 | 8.9 | |
| 44 | 6.9 | |
| 45 | 6.9 | |
| 46 | 6.9 | |
| 47 | 6.6 | |
| 48 | 6.6 | |
| 49 | 4.1 | |
| 50 | 3.6 |