Agentic Coding on SWE-bench Pro
52.6Pass@1DeepSeek-V4 Flash
Evaluation Results
| Method | Links | |
|---|---|---|
| DeepSeek-V4 Flash# Total Params=284B, # Active Params=13B2026.05 | 52.6 | |
| GPT-5.4 Nano2026.05 | 52.4 | |
| Qwen3.5# Total Params=397B, # Active Params=17B2026.05 | 50.9 | |
| Qwen3.6# Total Params=35B, # Active Params=3B2026.05 | 49.5 | |
| LAGUNA M.1# Total Params=225B, # Active Params=23B2026.05 | 49.2 | |
| LAGUNA XS.2# Total Params=33.4B, # Active Params=3B2026.05 | 46.3 | |
| Qwen3.5# Total Params=35B, # Active Params=3B2026.05 | 44.6 | |
| Claude Haiku 4.52026.05 | 39.5 | |
| Gemma 4# Total Params=31B, # Active Params=31B2026.05 | 35.7 |