METR says Anthropic's model progress is likely about twice as fast

METR says Anthropic was likely improving its models on well-specified tasks at about twice an earlier rate, as of October 2026. The nonprofit posted the assessment on November 5. The comparison is with the pace from March 2024 to October 2025. METR would not be surprised if the true rate were closer to 1.3 times or 3 times the earlier pace.

A separate benchmark line is more precise, and still a range. METR says scores on AECI, Anthropic's version of the Epoch Capabilities Index, have grown since October 2025 at an estimated 1.9 times the earlier rate. The 95 percent confidence interval is 1.5 to 2.8 times. The group says that implies about two years of earlier progress in one year. It does not name one model as the source of the speedup.

METR ties most of the current change to the use of AI, and says that is a judgment. It guesses that perhaps about 80 percent of the October 2026 acceleration comes from AI assistance rather than fully autonomous work. It says the late-2025 pickup was likely influenced by other factors, and that it is confident the change was not from faster growth in training compute. Staff surveys were consistent with a doubling, and METR says those surveys are hard to read.

Anthropic redacted a majority of the evidence METR found most useful. The review ran from August 12 to October 29, with access to internal training and benchmark data. METR says the public has to rely on its judgment more than it would like, and that it could be overstating the result if the measure fits only well-specified tasks. It also says it did not find a clear bottleneck that would stop further progress in the next 12 months, and that it is not confident in that reading.