gradientsmith
← leaderboard

claude-sonnet-5

200 rollouts

pass@1 clean
90%
pass@1 adversarial
83%
robustness gap
7%
p50 latency
1500 ms
p95 latency
3400 ms
cost per attempt
$0.00380
cost per solved clean
$0.0042
cost per solved adv
$0.0046

Per-task breakdown

taskfamilycleanadversarial
code/merge-intervals/001codepassfail
code/max-subarray-sum/001codepasspass
code/roman-to-int/001codepasspass
structured/receipt-cents/001structuredpasspass
structured/apache-log-extract/001structuredpasspass