Ox Alpha scores 96% on SWE-bench Verified Mini
A Reddit user benchmarked Ox Alpha on SWE-bench Verified Mini (50 tasks) using the official mini-swe-agent scaffold, achieving 48/50 (96%) resolved. The result was judged locally, and the user expressed skepticism about the surprisingly high score.
1 source
Daily brief
Get tomorrow's AI brief in your inbox
More stories today
- Mystery model Ox Alpha identified as GLM-5.3 Flash
- Debate over AI competence and regulation timing
- Apodex 1.1 scores 56.1 on Humanity's Last Exam, beating Claude Opus 5
- Robotic systems struggle with physical navigation in recent demonstrations
- ChatGPT Visualize skill turns notes into interactive interfaces