METR: Anthropic's model progress rate is about twice as fast as a year ago.

AI safety research group METR said Anthropic's model progress has roughly doubled in pace over the past year, with 11 weeks of its investigation indicating that advances made in the last year match what previously took about two years. METR attributed most of the acceleration, as of October, to AI itself, mainly AI-assisted human researchers rather than autonomous AI work, and said AI's contribution is growing. Employees using Anthropic's strongest internal models self-reported in an informal survey that they were about 12 times more productive than without AI assistance. Anthropic heavily redacted its public report, with co-founder Sam McCandlish citing commercially sensitive information and the risk of helping rivals speed up their own development. The findings matter as a signal that AI is increasingly compounding its own research velocity, and the next thing to watch is whether that acceleration shows up in future model releases and in less-redacted disclosures.