Terminal-Bench 2.0 on gpt-5-nano

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

What a task cost each configuration.

$ per solved task

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

Not reported: spoox-o-m

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

HarnessTokens per task

spoox-o-mReference1.2M×1.00

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

HarnessTime per task

spoox-o-mReference18.3 min×1.00

Up and to the left is better.

The frontier needs two configurations that report tokens, time or dollars.