SWE-Bench
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
SWE-bench Multilingual
21Pass@1
14
SWE-bench Lite
70GF Precision
14
SWE-Bench Pro 1.0 (test)
51.6Resolved Rate
14
SWE-Bench Multi
74Score
13
SWE-Bench Multilingual
71.7Accuracy
13
SWE-bench
70Exact Match (EM)
12
SWE-bench OpenHands
3.7Speedup
12
SWE-Bench Verified (held-out (sample of 150))
66Task Pass Rate
12
SWE-bench distinct-shard topology
34.2CF (median)
12
SWE-Bench-Verified (50 cases)
72Accuracy
12
SWE-Bench SWE-Agent-style workflows Verified
67.2Pass@1
12
SWE-bench Verified
76.2Success Rate
12
SWE-bench Lite Oracle (test)
25.83Accuracy
12
SWE-bench Verified single-run, single-patch evaluation protocol (test)
51.6Resolution Rate
11
SWE-BENCH LIVE
24.7Resolution Rate
11
SWE-bench Verified
73.8Score
10
SWE-bench Verified
78Average Success Rate @1
10
SWE-Bench Verified
86.2Resolution Rate (%)
10
SWE-Bench Lite
53.7Pass@1
10
SWE-bench Multilingual
77.5Score
10
SWE-Bench Lite
24Solve Rate
9
SWE-bench
0.45Similarity Score
9
SWE-bench Pro
52.6Pass@1
9
SWE-bench In-Distribution
52.34Resolved Rate
9
SWE-bench
65Test Success Rate (TSR)
9