Anthropic wants you to know Claude leads a quarter of its research, but "lead" doesn't mean what you think
Anthropic says Claude now leads 26% of its research tasks, a Claude-scored metric with ambiguous definitions and minimal external verification.
Anthropic published metrics claiming Claude reaches Epoch AI's AL4 ("AI leads") level on 26% of its research work, up from under 1% in February 2026, but the scores were assigned by a Claude "judge" model that matched human self-ratings only 59% of the time. About 30,000 agents run concurrently on Anthropic's most-used internal platform, with real-time monitoring blocking just 0.002% of more than a billion decisions in August. The company also disclosed that roughly 6% of AI research compute went to safety work during a sample week in July, arguing compute share could serve as an industry slowdown lever.
- 26% of Anthropic research work now sits at Epoch AI's AL4 'leads' level, up from under 1% in February 2026
- A Claude judge model assigned the scores, matching human self-ratings only 59% of the time
- About 30,000 agents run concurrently internally; real-time monitoring blocked 0.002% of a billion-plus August decisions
- Roughly 6% of AI research compute went to safety work in a sample July week
- Critics may read the 'leads' framing as supporting Amodei's call for coordinated frontier slowdown
From the original article
3 min read at the-decoder.com
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.
By Maximilian Schreiner · the-decoder.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.