The reported benchmark is a useful but limited empirical check that weakens the pelicanmaxxing hypothesis without establishing anything about training practices across AI labs.
Limits of the evidence: The source is a discovery-class summary of the analysis, the underlying methodology and results were not independently verified here, and the sample covers only seven named models and a narrow prompt set.
What to watch
Taken from the first report; the whole story has not been re-assessed yet.
Are AI labs pelicanmaxxing?
Simon WillisonJul 22, 2026First report
No revisions yet.
No maturity ladder applies to this desk.
Sens.ai aggregates and assesses published reporting. The assessment above is machine generated; the original sources are authoritative.