KernelBench-Mega is the Mega track of kernelbench.com/mega, an independent agentic GPU-kernel benchmark built by Elliot Arledge; the site states it is not affiliated with Stanford’s KernelBench and publishes its source (Infatoshi/kernelbench.com). The track asks a model to fuse a whole decode step (its listed task is a Kimi-Linear decode) into one launch, and entries can be flagged and rejected on audit rather than counted. In this corpus it appears once: Claude Fable 5’s 18.7x megakernel of July 2026, which the corpus files in the same kernel-writing strand as GPT-5.5 topping Stanford’s KernelBench at 6.57% in April, though the two benchmarks are unrelated, under silicon-designs-itself.
Benchmark
KernelBench-Mega
The megakernel track of Elliot Arledge's independent kernelbench.com agentic GPU-kernel benchmark (not affiliated with Stanford's KernelBench), on which Claude Fable 5 wrote the first genuine single-launch decode megakernel.
Sources
Attribution and reporting linked by this record; links do not imply independent verification.
- Welcome to July 3, 2026 (theinnermostloop.substack.com)
Development timeline
Newest reporting dates first. Reporting dates may differ from event dates.
Knowledge graphopen in browser →
Published by
Cite or correct this record
Suggest a correction — opens a GitHub draft for you to review and submit.