Claude 3.5 hits 49% on SWE-bench: AI fixes real codebases
Posted by Chaitanya Mahaprabhu · 2025-02-22
Claude 3.5 resolves 49% of SWE-bench issues — real bugs in real codebases, fixed autonomously. We're past the toy phase. AI agents that write production code, run tests, and submit PRs are here.