Big Pickle on SWE Atlas – Codebase QnA

The big-pickle model achieved a 50.8% task resolve rate on the Scale AI SWE Atlas Codebase QnA benchmark using the mini-swe-agent scaffold. Within the same scaffold class, it outperformed GLM 5.2 and GPT-5.6-Sol, trailing only Claude-based models running on the native Claude Code scaffold.