3 ms·
#1 open-source agent on SWE-Bench Verified by combining Claude 3.7 and O1
- nuatsimon 2y agohuge props to the team for open sourcing this
- colinflaherty 2y agoColin here, author of the post - would love to answer questions about this. And make sure to try out the open-source repo! It's a super easy starting point for experimenting with coding agents. It's nearly one-click to run agents in isolated Docker containers on SWE-bench Verified problems, ensemble the results, and run the SWE-bench evaluation harness to compute scores. Check it out here: https://github.com/augmentcode/augment-swebench-agent https://github.com/augmentcode/augment-swebench-agent
- arunchaganty 2y agoNice! I know it's super hard to sota a benchmark, especially a few years into it, so congrats on the milestone!