Skip to content
Hbench
hbench

humaneval-plus@1.0 (unreviewed) on GPT-5.6 Luna

164 tasks, 1 attempt, open network. Updated 24 Oct 2026.

164 tasks from humaneval-plus@1.0 on ghcr.io. Nobody has reviewed this dataset, so a score here says nothing about the benchmark it's named after.

Build a run for this board

Value

What a task cost each configuration. Claude Code used 23% more tokens than Codex, and 97% more time per task.

$ per solved task

Lower is cheaper. Dollars per task divided by the pass rate.

Embed this chart

Harness$ per solved task
Codex0.164.0 · Reference$0.003$0.003 a task
Claude Code2.1.309$0.10$0.10 a task

Tokens per task

Lower is leaner. Input plus output, cached input included.

Embed this chart

HarnessTokens per task
Codex0.164.0 · Reference48k×1.00
Claude Code2.1.30959k×1.23

Agent time per task

Lower is faster. Agent start to finish; setup and tests left out.

Embed this chart

HarnessTime per task
Codex0.164.0 · Reference17 s×1.00
Claude Code2.1.30933 s×1.97

Pass rate against tokens per task

Up and to the left is better. Pass-rate axis starts at 85%.

Embed this chart

85%90%95%100%45k50k55k60k65kTokens per task, input and outputMore tokens for a lower pass rate than CodexCodex0.164.0 · 48k · 95.7%Claude Code2.1.309 · 59k · 93.9%
  • On the frontier: no other row has a higher pass rate for less tokens
  • Off the frontier
  • 95% interval of the pass rate
On the frontier, least tokens first: Codex 0.164.0, 48k tokens a task, 95.7%.
  • 2 runs
  • Updated 24 Oct 2026
  • Method