Benchmarks / Coding
Aider Polyglot
Evaluates LLMs on 225 Exercism coding exercises across six languages, testing code editing without human intervention. Maintained by Paul Gauthier. Score: Percent of the 225 Exercism exercises completed with a correct edit.
Aider Polyglot
Higher is better · best effort per model
full table below
| # | Model | Effort | Score |
|---|---|---|---|
| 1 | GPT-5 | high | 88.0% |
| 2 | o3 pro | high | 84.9% |
| 3 | Gemini 2.5 Pro | high | 83.1% |
| 4 | o3 | high | 81.3% |
| 5 | Grok 4 | high | 79.6% |
| 6 | GPT-4.1 | high | 78.2% |
| 7 | DeepSeek V3.2 | high | 74.2% |
| 8 | Opus 4 | medium | 72.0% |
| 9 | o4 mini | high | 72.0% |
| 10 | DeepSeek R1 | xhigh | 71.4% |
| 11 | o1 | high | 61.7% |
| 12 | Sonnet 4 | high | 61.3% |
| 13 | o3 mini | high | 60.4% |
| 14 | Qwen 3 235B-A22B | xhigh | 59.6% |
| 15 | Kimi K2 | medium | 59.1% |
| 16 | Gemini 2.5 Flash | medium | 55.1% |
| 17 | DeepSeek V3 | none | 55.1% |
| 18 | Grok 3 | none | 53.3% |
| 19 | Sonnet 3.5 | medium | 51.6% |
| 20 | Grok 3 Mini | high | 49.3% |
| 21 | GPT-4.5 | none | 44.9% |
| 22 | gpt-oss-120b | high | 41.8% |
| 23 | Qwen 3 32B | none | 40.0% |
| 24 | Gemini 1206 | – | 38.2% |
| 25 | Gemini 2.0 Pro | none | 35.6% |
| 26 | o1 mini | high | 32.9% |
| 27 | GPT-4.1 Mini | high | 32.4% |
| 28 | Haiku 3.5 | medium | 28.0% |
| 29 | Qwen 2.5 Coder | none | 26.2% |
| 30 | GPT-4o | high | 23.1% |
| 31 | Gemini 2.0 Flash | medium | 22.2% |
| 32 | DeepSeek V2.5 | none | 17.8% |
| 33 | Llama 4 Maverick | medium | 15.6% |
| 34 | Codestral 25.01 | – | 11.1% |
| 35 | GPT-4.1 Nano | high | 8.9% |
| 36 | Gemma 3 27B IT | medium | 4.9% |
| 37 | GPT-4o mini | high | 3.6% |