Announcement_20
🎉 With test-time scaling (TestGRAD), which mines failure patterns from test attempts to pick the right patch, our 4-agent ensemble reached 84.2% on SWE-bench Verified (as of May 2026) 🚀
🎉 With test-time scaling (TestGRAD), which mines failure patterns from test attempts to pick the right patch, our 4-agent ensemble reached 84.2% on SWE-bench Verified (as of May 2026) 🚀