Skip to content
Hbench
hbench

SWE-bench Verified on Claude Opus 5.5

500 tasks, 1 attempt, allowlisted network. Updated 19 Oct 2026.

Real GitHub issues, resolved against the projects' own test suites. Comparable with published full-set numbers, but scores this high on a public set may reflect contamination: read them as harness differences on one model, not as the model's skill. About 5-7 hours and about $80 per run. Bring your own Anthropic key.

Build a run for this board

Ranking

Pass rate by harness, higher is better

Embed this chart

Δ vs DeepSeek Harness, pts
DeepSeek Harness0.2.0-rc.293.4%95% interval 91.2–95.4DeepSeek Harness 0.2.0-rc.2 is the reference.
Claude Code2.1.30290.8%95% interval 88.2–93.2Claude Code 2.1.302 minus DeepSeek Harness 0.2.0-rc.2: −2.6 points, 95% interval −4.6 to −0.8, over 500 tasks both ran; a clear difference.

All harnesses

Reading this table

Every harness on this board, in its order
HarnessRankLast published
DeepSeek Harness0.2.0-rc.2119 Oct 2026run #17
Claude Code2.1.302219 Oct 2026run #18