BLXBench Refactoring category
Code · 2026-04-25
Refactoring category component of the BLXBench public leaderboard. The source describes this category as code transformation while preserving behavior and intent; scores are category score percentages.
Top models (higher is better)
| Model | Score |
|---|---|
| North Mini Code | 69.1 |
| GPT-5.3-Codex | 67.1 |
| Qwen3.7-Max | 66.7 |
| MiMo-V2.5 | 66.0 |
| GPT-5.5 | 65.2 |
| Nemotron 3 Super 120B A12B | 64.7 |
| GLM-5.1 | 64.3 |
| Granite 4.1 8B | 63.4 |
| MiniMax M3 | 62.5 |
| Mistral Small 4 | 62.5 |
| DeepSeek-V4-Pro | 61.9 |
| Mistral Medium 3.5 | 61.2 |
| DeepSeek-V4-Flash | 61.0 |
| Grok Build 0.1 | 60.1 |
| GLM-5.2 | 59.9 |
| Nemotron 3 Nano 30B A3B | 59.5 |
| Kimi K2.6 | 59.3 |
| Qwen3.6-Flash-2026-04-16 | 59.0 |
| MiMo-V2.5-Pro | 55.8 |
| Grok 4.3 | 55.6 |
| Qwen3.7-Plus | 55.1 |
| Ring 2.6 1T | 54.7 |
| CoBuddy | 51.0 |
| Nemotron 3 Nano Omni 30B A3B | 50.8 |
| Opus 4.8 | 46.4 |
| Gemini 3.1 Flash-Lite | 43.8 |
| Claude Fable 5 | 42.5 |
| Kimi K2.7 Code | 37.3 |
| Opus 4.7 | 36.8 |
| MiniMax M2.7 | 28.8 |
| Gemini 3.5 Flash | 25.0 |
| Step 3.7 Flash | 17.7 |