Inference
EN
SWE-bench team

SWE-bench (full test split)

Resolved-issue rate on the full 2,294-issue SWE-bench test split, which the leaderboard site does not publish a board for.

ModelsModels with a result.7
Top resultBest result on this benchmark.52.6%Claude Opus 4.5
Top-3 spreadPoints from first to third.31.1 pts
UpdatedDate the source changed the results.19 Dec 2025

Results

7 results
#ModelResultThe score from the source.
01Claude Opus 4.5Sonar Foundation Agent52.6%
02Claude 3.7 SonnetSWE-agent 1.033.8%
03Claude 3.5 SonnetSWE-agent21.5%
04GPT-4oAppMap Navie14.6%
05Claude 3 OpusSWE-agent7.1%
06Claude 2RAG baseline2.0%
07SWE LlamaRAG baseline0.7%