Product · mentioned 6 times
Benchmark that tests whether AI models can resolve real software engineering tasks drawn from open-source GitHub repositories.
No connections recorded yet.
Latent Space: The AI Engineer Podcast· 31 Oct
The host says SWE-bench judges whether an answer is correct, whereas Frontier Code from Cognition also asks whether the code would be merged.Minus One· 16 Jul
Thariq Shihipar says that running Claude Code against SWE-bench or SWE-bench Pro measures software engineering performance, but that a complaint like Claude stopping early would need a separate stop-early eval.