FrontierSeptember 18, 2026via The Decoder
Anthropic wants you to know Claude leads a quarter of its research, but "lead" doesn't mean what you think
Why it matters
Anthropic is using self-serving metrics to claim AI progress on research productivity, but the underlying methodology is opaque enough to warrant skepticism. This matters for how practitioners and investors should weight Anthropic's capability claims and self-sufficiency narrative.
Key signals
- Claude reported to 'lead' 26% of Anthropic's future model research work
- Up from under 1% in February 2026
- Metrics are self-reported by Claude, not independently audited
- Definition of 'lead' appears to be narrower than headline suggests
- Scale of measurement is unclear/fuzzy
- First time Anthropic has published such metrics
- Claude now 'leads' 26% of work on future models, up from <1% in February
- Metrics are self-generated (Claude scores its own contribution)
- 'Lead' is undefined and may not mean what Anthropic implies
- Underlying scale and scoring methodology are not disclosed
- First public release of internal build metrics by Anthropic
The hook
Anthropic claims Claude now 'leads' 26% of its own model research — but the metric is self-measured, the baseline fuzzy, and 'lead' means something much smaller than the headline.
For the first time, Anthropic is releasing metrics on how it builds its own AI. Claude already "leads" 26 percent of the work on future models, up from under one percent in February. But the underlying scale is fuzzy, the scoring comes from Claude itself, and "lead" means less than it sounds.