Atlas

Benchmarks

KernelBench Mega

KernelBench Mega evaluates agent-generated kernels on a two-problem GPU optimization deck. This suite-level row defines the family and has no direct score observations; hardware-specific child benchmarks report the best valid Kimi-Linear decode speedup.

Models
0
Highest score
—
Highest
—
Updated
—
Resolution
—
Saturation
—

No signal or difficulty estimate. Held out of the fit by design.

This is a benchmark-family page. Scores are recorded on the component benchmarks below rather than directly on the parent, because the components use different score scales and directions.

MetricModelsLeaderDirectionDate
KernelBench Mega — RTX PRO 6000 Best Decode Speedup12Claude Fable 518.7higher is better2026-07-17
KernelBench Mega — B200 Best Decode Speedup6Opus 4.819.4higher is better2026-07-17
KernelBench Mega — H100 Best Decode Speedup5Opus 4.815.5higher is better2026-07-17
Loading Atlas data…