evals.report
BenchmarksLabsCompareRun guidesIn the wild
Models
1 selected
DeepSeek V4 ProDeepSeek
Benchmarks
3 selected
SWE-bench VerifiedCodingDeepSWECodingSWE-bench ProCoding
BenchmarkDeepSeek V4 ProOpenDeepSeek
SWE-bench Verified% resolved80.6%
DeepSWE% resolved7.52%
SWE-bench Pro% resolved55.4%

No aggregate score is calculated. Each row uses its benchmark’s own metric. Compare rows independently.