Anthropic is likely improving its models on well-specified tasks at about twice the rate it managed between March 2024 and October 2025 — and most of that acceleration appears to come from AI helping to build AI. That is the central finding of an independent assessment by METR published on November 5, 2026. Why it matters: it is some of the clearest evidence yet that AI is speeding up its own development, which changes how fast capabilities — and risks — can move.
November 6, 2026
What happened
METR, the nonprofit also known as Model Evaluation and Threat Research, spent 11 weeks inside Anthropic’s data, from August 12 to October 29. Its team reviewed internal benchmarks, training and research information, and interviewed about 15 staff members. METR said it took no payment from Anthropic, which provided laptops and free tokens.
Using AECI — Anthropic’s version of the Epoch Capabilities Index, which combines benchmark results into one score — METR estimated progress since October 2025 has run at 1.9 times the earlier rate (95% confidence interval: 1.5 to 2.8 times): roughly two years of historical progress in about one year.
The details that matter
METR was explicit that the estimate is highly uncertain. It said it would not be surprised if the true acceleration were closer to 1.3 times or 3 times the historical rate, and Reuters reported METR putting the plausible range at 1.5 to 2.5 times in its summary comments. METR also said it is confident the acceleration was not caused by faster growth in training compute.
On cause, METR’s judgment is that AI accounts for a majority of the acceleration — perhaps about 80% as of October 2026 — while other factors played a large role in late 2025. It also said AI cannot yet make highly significant, fully autonomous contributions to the most important stages of model training.
There is a major transparency caveat: Anthropic held unilateral redaction rights, and METR said it redacted a majority of the evidence METR found most informative. Anthropic described the redactions as removing information that could accelerate other developers.
Why it matters
For readers, the takeaway is not that today’s models are suddenly twice as capable — METR measured the rate of progress, not a doubling of capability. The practical implication is timing: if the rate holds, improvements that used to take two years could arrive in one, and evaluation and safety work has less time to keep up. That is the “everything AI, tested” angle worth watching — independent measurement, with its limits stated plainly, rather than a vendor’s own claim.
Frequently asked questions
What does a 1.9x acceleration mean?
It means the capability index METR tracked has been rising 1.9 times faster since October 2025 than it did in the earlier period — roughly two years of historical progress compressed into one year.
Did AI cause the acceleration?
METR’s judgment is that AI likely drove a majority of it, perhaps about 80%, but it calls that estimate substantially uncertain and notes other factors contributed, especially in late 2025.
Can the findings be fully checked?
No. Anthropic redacted a majority of the evidence METR found most informative, so outside readers cannot independently verify the strongest supporting material.
Sources: Reuters via SRN News, METR assessment reporting via AI Stock Wire and other outlets summarizing METR’s November 5 assessment.