SWE-Bench Pro (Public Dataset)
Code · 2025-09-19
Percent of challenging long-horizon software engineering tasks resolved in public repositories.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Fable 5 | 80.3 |
| Claude Mythos 5 | 80.3 |
| Claude Opus 5 | 79.2 |
| Claude Mythos Preview | 77.8 |
| Opus 4.8 | 69.2 |
| GPT-5.6 Sol | 64.6 |
| Opus 4.7 | 64.3 |
| Sonnet 5 | 63.2 |
| GPT-5.6 Luna | 62.7 |
| Muse Spark 1.1 | 61.5 |
| Qwen3.7-Max | 60.6 |
| Laguna S 2.1 | 59.4 |
| Gemini 3.6 Flash | 58.7 |
| GPT-5.5 | 58.6 |
| Hy3 | 57.9 |
| GPT-5.4 | 57.7 |
| Opus 4.6 | 56.7 |
| GPT-5.2-Codex | 56.4 |
| MiniMax M2.7 | 56.2 |
| MiMo-V2.5 | 56.1 |
| GPT-5.3-Codex | 56.0 |
| Inkling-Small | 55.9 |
| GPT-5.2 | 55.6 |
| DeepSeek-V4-Pro | 55.4 |
| Muse Spark | 55.0 |
| GPT-5.4 Mini | 54.4 |
| Inkling | 54.3 |
| Gemini 3.1 Pro Preview | 54.2 |
| Gemini 3.5 Flash-Lite | 54.2 |
| Sonnet 4.6 | 53.8 |
| Inkling-Small Preview | 53.2 |
| DeepSeek-V4-Flash | 52.6 |
| GPT-5.4 Nano | 52.4 |
| Opus 4.5 | 52.0 |
| GLM-5.2 | 51.8 |
| GPT-5.3-Codex-Spark | 51.5 |
| GPT-5.6 Terra | 51.3 |
| Qwen3.5 397B A17B | 50.9 |
| Kimi K2.6 | 50.1 |
| GLM-5.1 | 49.2 |