Terminal-Bench 4.0.0 · Opus 5 / Claude Code · reasoning effort low

Created by Ryan Marten
||Owned by Terminal-Bench
PublicRerun job
330 / 330 finished
Started Sep 16, 2026, 8:08 PM · Ran for 9h 20m
326 completed0 timeouts4 errors0 in progress0 pending
Average reward
0.35
Across scored trials
Total tokens
2,721,178,403
8,245,995 tokens/trial · 97% cache hit
Inference cost
$2,393.88
$7.25 per finished trial
Sandbox cost
—

Results

Metric: reward
Group by0 groups
Loading results…