Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

SWE-Bench

Benchmarks

Task NameDataset NameSOTA ResultTrend
Software Engineering Issue ResolutionSWE-bench Verified
Resolution Rate89.2
110
Agentic CodingSWE-bench Verified
Percentage Resolved87.6
71
Software Engineering Task ResolutionSWE-bench Verified
Resolution Rate73.3
63
Software EngineeringSWE-bench Verified
Success Rate80.9
53
Software EngineeringSWE-bench Verified
Accuracy62.6
43
Automated Software EngineeringSWE-bench Verified
Resolved Rate1,770
39
Software EngineeringSWE-bench Lite
Speedup4.66
36
Issue ResolutionSWE-bench Verified (test)
Pass Rate77.2
36
Software EngineeringSWE-bench verified (All)
Success Rate93.8
32
Software EngineeringSWE-bench Verified
Resolution Rate83.8
32
Software Engineering Agent TaskSWE-Bench Pro
Pass@3100
28
Function-level Code LocalizationSWE-bench Live Lite
Acc@174.8
25
File-level Code LocalizationSWE-bench Live Lite
Acc@182.1
25
Function-level Code LocalizationSWE-bench Verified (Lite)
Acc@183.4
25
File-level Code LocalizationSWE-bench Verified Lite
Accuracy@191.9
25
Code GenerationSWE-Bench
Speedup Ratio2.96
24
Code LocalizationSWE-bench Verified (test)
File Precision86.38
24
Software EngineeringSWE-Bench Lite
Accuracy57.09
23
Software EngineeringSWE-bench Verified
Pass@172
23
File-level localizationSWE-Bench Verified
Hit@189.6
21
Software Engineering Issue ResolutionSWE-Bench-Pro (test)
Pass@159.9
21
Code AgentSWE-Bench Verified
Score0.809
21
Agentic CodingSWE-bench 50 tasks sampled
Resolved Count12
20
Software EngineeringSWE-Bench Verified
Pass Rate72
20
Software EngineeringSWE-Bench Multilingual 1.0 (test)
Resolution Rate75.2
20
Showing 25 of 160 rows