Code Migration
Code · 2026-06-16
Code-migration benchmark evaluating whether language models can reimplement real-world programs in a different programming language while preserving behavior.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Opus 5 | 57.5 |
| Claude Fable 5 | 55.1 |
| GPT-5.6 Sol | 52.9 |
| Opus 4.8 | 47.3 |
| GPT-5.5 | 45.2 |
| GPT-5.6 Luna | 44.5 |
| Sonnet 5 | 44.4 |
| Opus 4.7 | 43.9 |
| Sonnet 4.6 | 39.9 |
| GLM-5.2 | 37.9 |
| Grok 4.5 | 36.6 |
| GPT-5.6 Terra | 36.4 |
| GPT-5.4 | 35.0 |
| Muse Spark 1.1 | 31.1 |
| Gemini 3.6 Flash | 30.9 |
| Kimi K2.6 | 27.8 |
| Gemini 3.5 Flash | 26.7 |
| DeepSeek-V4-Pro | 26.2 |
| GLM-5.1 | 25.8 |
| Kimi K2.7 Code | 25.4 |
| MiMo-V2.5-Pro | 21.6 |
| MiniMax M3 | 19.9 |
| Gemini 3.1 Pro Preview | 17.3 |
| MiMo-V2.5 | 14.2 |
| GPT-5.4 Mini | 12.9 |
| Inkling | 11.8 |
| Qwen3.6 Plus (2026-04-02) | 11.1 |
| MiniMax M2.7 | 8.7 |
| Haiku 4.5 | 7.6 |
| Grok 4.3 | 6.8 |
| Gemini 3 Flash Preview | 6.4 |
| Gemini 3.5 Flash-Lite | 6.1 |
| Gemini 3.1 Flash-Lite Preview | 4.6 |