Terminal-Bench 2.1 · GPT-6 Astra / Codex · reasoning efforts low-max

Created by Ryan Marten
||Owned by Terminal-Bench
PublicRerun job
2,225 / 2,225 finished
Started Aug 31, 2026, 8:10 PM · Ran for 21h 16m
2,160 completed14 timeouts51 errors0 in progress0 pending
Average reward
0.87
Across scored trials
Total tokens
1,762,173,468
791,988 tokens/trial · 94% cache hit
Inference cost
$3,814.32
$1.71 per finished trial
Sandbox cost
—

Results

Metric: reward
Group by0 groups
Loading results…

0 groups

Terminal-Bench 2.1 · GPT-6 Astra / Codex · reasoning efforts low-max | Harbor