Anthropic’s use of AI in its own research is speeding the development of stronger models, according to a new independent assessment that puts its recent pace of benchmark progress at about twice its earlier rate.
The assessment by METR, an independent AI research nonprofit, ran from August 12 to October 29 with unusual access to Anthropic’s internal data on its research process, including training metrics, benchmark scores for internal models and interviews with staff. The report was published Thursday.
METR found that benchmark scores for Anthropic’s models have been growing at an estimated 1.9 times the rate seen before October 2025, within a wide uncertainty range of 1.5 to 2.8 times, implying roughly two years of its historical progress in the past year. The assessors said they are confident the acceleration was not driven by faster growth in training compute.
A majority of the acceleration is likely attributable to Anthropic’s extensive use of AI to do research and development, METR said, while stressing significant uncertainty and that other causes likely played a large role late last year. Anthropic staff described AI accelerating research mainly by assisting human researchers, not through fully autonomous contributions, so the findings do not show self-improving AI or anything approaching AGI.
METR also said Anthropic redacted a majority of the evidence it found most informative, which Anthropic described as removing confidential information that could accelerate other AI developers. That makes it hard for the public to judge the evidence, METR said, and it noted the acceleration could also be overstated if non-AI factors or quirks of the benchmark explain part of it.
Even so, the assessors warned that because they saw no significant bottlenecks ahead, there is a significant risk that progress could accelerate further over the next year.
JBizNews Desk | San Francisco, California
© JBizNews.com. All rights reserved. Unauthorized reproduction or redistribution is strictly prohibited.
.jpg){kind=link}