Skip to content
Hbench

Claude Code

claude-code

Claude Code's pass rate on every board it ran on. Scores on different boards don't add up, so this page doesn't add them.

On each board

Terminal-Bench 2.0

terminal-bench@2.0
GLM-4.72.1.34, full, 89 tasks, 5 attempts, open
33.3%25.2–41.6

SWE-bench Verified

swebench-verified@1.0
Claude Opus 5.52.1.302, full, 500 tasks, 1 attempt, allowlist
Rank 2
2 harnesses
90.8%88.2–93.2Claude Code 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.

humaneval-plus@1.0 (unreviewed)

Every board here: 2.1.309, full, 164 tasks, 1 attempt, open
Ranks 1–2
2 harnesses
95.1%91.5–98.2Claude Code 2.1.309 is the reference.
Ranks 1–2
2 harnesses
93.3%89.0–97.0Claude Code minus Codex 0.164.0: −1.2 points, 95% interval −4.3 to +1.8, over 164 tasks both ran; too close to call.
Ranks 1–2
2 harnesses
93.9%90.2–97.6Claude Code minus Codex 0.164.0: −1.8 points, 95% interval −4.9 to +0.6, over 164 tasks both ran; too close to call.
95.7%92.7–98.2
Ranks 1–2
2 harnesses
93.9%90.2–97.0Claude Code minus Codex 0.164.0: −1.8 points, 95% interval −4.9 to +0.6, over 164 tasks both ran; too close to call.
Ranks 1–3
3 harnesses
95.7%92.7–98.2Claude Code minus DeepSeek Harness 0.2.1-rc.2: −0.6 points, 95% interval −1.8 to 0.0, over 164 tasks both ran; too close to call.
Ranks 1–2
2 harnesses
96.3%93.3–98.8Claude Code 2.1.309 is the reference.
Ranks 1–2
2 harnesses
92.7%88.4–96.3Claude Code 2.1.309 is the reference.