Skip to content
Hbench
hbench

Terminal-Bench 2.0 on GPT-5.5

89 tasks, 5 attempts, open network. Updated 25 Oct 2026.

Build a run for this board

Value

What a task cost each configuration.

$ per solved task

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

Embed this chart

Not reported: Capy, clnkr

Tokens per task

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

Embed this chart

Not reported: Capy, clnkr

Agent time per task

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

Embed this chart

HarnessTime per task
clnkr0.3.116.4 min×0.80
CapyReference8.0 min×1.00

Pass rate against time per task

Up and to the left is better. Not offered: tokens and dollars, which fewer than two rows report. Pass-rate axis starts at 55%.

Embed this chart

55%65%75%85%95%5.8 min6.7 min7.5 min8.3 minAgent time per taskMore time for a lower pass rate than a row on the frontierCapyversion not recorded · 8.0 min · 83.1%clnkr0.3.11 · 6.4 min · 66.1%
  • On the frontier: no other row has a higher pass rate for less time
  • Off the frontier
  • 95% interval of the pass rate
On the frontier, least time first: clnkr 0.3.11, 6.4 min a task, 66.1%; Capy, 8.0 min a task, 83.1%.
  • 2 runs
  • Updated 25 Oct 2026
  • Method