Inference
EN
SWE-bench team

SWE-bench Multilingual · swebench.com

Resolved-issue rate on real GitHub issues, as published by the SWE-bench team.

ModelsModels with a result.12
Top resultBest result on this benchmark.72.7%Gemini 3 Flash
Top-3 spreadPoints from first to third.2.0 pts
UpdatedDate the source changed the results.20 Feb 2026

Result and cost

12 of 12 results

Results

12 results
#ModelResultThe score from the source.CostUS dollars to run one task.
01Gemini 3 Flashmini-SWE-agent72.7%$0.352
02Claude 4.6 Opusmini-SWE-agent72.0%$0.663
03Claude 4.5 Opusmini-SWE-agent70.7%$0.834
04GLM-5mini-SWE-agent69.7%$0.642
05Gemini 3 Promini-SWE-agent68.7%$1.02
06MiniMax M2.5mini-SWE-agent68.3%$0.101
07Kimi K2.5mini-SWE-agent67.3%$0.693
08Claude 4.5 Sonnetmini-SWE-agent67.0%$0.67
09GPT-5.2high effort · mini-SWE-agent66.7%$0.536
10Claude 4.5 Haikumini-SWE-agent64.7%$0.379
11DeepSeek V3.2mini-SWE-agent59.0%$0.384
12GPT-5 minimini-SWE-agent39.7%$0.052