evals.report
BenchmarksLabsCompareRun guides

SWE-Lancer

An OpenAI benchmark of over 1,400 real freelance software engineering tasks from Upwork (worth $1M in total payouts), where models either implement IC SWE code patches graded by end-to-end tests or act as SWE managers selecting the best technical proposal, measured by task pass rate and dollars earned.

CodingIC SWE pass rate (Diamond)Higher is better
ModelLabScoreSource modelStatusDate
Claude 3.5 SonnetAnthropic26.2%VerifiedJun 20, 2024Details
GPT-4oOpenAI8.6%VerifiedMay 13, 2024Details

Each row reports the model’s IC SWE pass rate (Diamond) on SWE-Lancer. Click a row for the full run context.