Anthropic now improves its AI about twice as fast as before, METR estimates
The nonprofit METR reviewed Anthropic's internal data for about 11 weeks and says much of the speed-up comes from AI helping to build AI.
In brief
Anthropic's AI progress is about 2x faster
METR reviewed Anthropic's internal data for about 11 weeks. It estimates a pace twice that of 2024-2025, with a range of roughly 1.3x to 3x. Anthropic redacted most of the key evidence.
The independent nonprofit METR said on November 5, 2026 that Anthropic is likely improving its AI models about twice as fast as it did between March 2024 and October 2025. The claim comes from a public assessment based on internal company data and staff interviews.
The estimate covers "well-specified tasks", meaning jobs with clear goals. METR gave it with high uncertainty. It said the true figure could be closer to 1.3 times or 3 times the earlier pace.
AI is speeding up our research, but it's hard to say exactly how much.
How the review worked
METR staff ran a high-access evaluation at Anthropic from August 12 to October 29, 2026, about 11 weeks. They saw data on training, benchmark scores for internal models and analyses of AI speed-ups. They also interviewed staff. METR called the access unprecedented.
Why AI itself matters here
METR said most of today's acceleration is likely due to Anthropic's heavy use of AI in its own research. It added that other causes probably played a large role in late 2025. Anthropic has said its Claude chatbot increasingly helps build newer versions of itself.
METR's concern is practical. If AI began to speed up AI development significantly, extreme and risky capabilities could appear suddenly.
The timing
The findings landed while CEO Dario Amodei is calling for the industry to slow down. In a September 12 essay he urged it to slow the pace at which AI capabilities improve. Anthropic is also reportedly planning a stock market debut later this month.
Co-founder Sam McCandlish replied on X that many of METR's conclusions match his own. He wrote that AI is speeding up their research, but it is hard to say by how much. He said Anthropic will publish its own analysis in the coming weeks.
Where the numbers differ
The Associated Press framed the finding as twice the pace of a year earlier. METR's own baseline is the longer March 2024 to October 2025 period. Other analyses reported by outlets give different ranges: a robustness check of about 1.1 to 2.9 times, and a benchmark-based estimate of 1.9 times with a 95% interval of 1.5 to 2.8.
Attribution also varies. METR's tentative view puts about 80% of the acceleration on AI use. Another analysis put it at only 15 to 20%.
Anthropic redacted most of the evidence METR found most informative, saying it removed confidential information that could help other AI developers. METR said the public must therefore lean on its judgment, and that Anthropic's public risk report understates the acceleration. Anthropic did not immediately comment on the initial reports.