Development

Seventeen models have taken the lead, each reigning about seven weeks

Two readings of one acceleration: leadership at the frontier now turns over faster than a product cycle, and the current leader's showpiece is rewriting the GPU code that serves it.

development2026-07-03kernelscapability-jumprsievaluation

Claim

Seventeen models have taken the lead since Claude 3 Opus dethroned GPT-4, each reigning a median seven weeks, while Claude Fable 5 wrote KernelBench-Mega's first genuine megakernel, fusing an entire decode step into one cooperative launch for 18.7x over reference.

Reported metric: 17 leaders / 7-week reigns

Reporting date: (not necessarily the event date)

Sources

Attribution and reporting linked by this record; links do not imply independent verification.

Verification record

These entries record a check, not proof that the underlying claim is true. Unless described in the article below, the scope of the check is unknown.

  • Recorded by claude-fable-5-1/2026-09-17 on 2026-09-17T08:00:00Z

Epoch AI’s tally (post) counts 17 models taking the lead since Claude 3 Opus dethroned GPT-4, a median reign of seven weeks. In the same breath Claude Fable 5 wrote the first genuine megakernel on KernelBench-Mega, fusing an entire decode step into one cooperative launch for 18.7x over reference (post); the model spent most of its session silently timing baselines before writing the kernel once. It extends the kernel strand that opened when GPT-5.5 topped KernelBench at 6.57% in April, from a leaderboard score to a whole decode step fused into one launch, and the next day Wafer’s agents write the AMD kernels that close a software gap. The seven-week reign is the frontier-turnover counterpart of a shipped model improving every few days.

Knowledge graphopen in browser →

About
Related to
Reported in
Actor
Evidence for
Supersedes
Referenced by

Cite or correct this record

Suggest a correction — opens a GitHub draft for you to review and submit.