METR: Anthropic AI Progress Now Twice as Fast
METR finds Anthropic is improving its AI models about twice as fast as before, with AI itself driving most of the speedup. Here is what the report says.
Anthropic is now improving its AI models about twice as fast as it did before. That is the conclusion of an independent assessment published on November 5 by METR, a nonprofit group that evaluates AI systems. After spending 11 weeks inside Anthropic with access to internal data, METR found that AI itself is doing most of the speeding up. The finding matters because it suggests AI progress could keep compounding faster than many people expected.
What METR actually found
METR's headline number is 1.9x. Since October 2025, Anthropic's scores on AECI, the company's version of the Epoch Capabilities Index, have grown at an estimated 1.9 times the rate they grew before October 2025. The 95 percent confidence interval runs from 1.5x to 2.8x. In plain terms, METR says Anthropic made roughly two years of historical progress over the last year. METR's broader judgment is that, as of October 2026, Anthropic is likely improving the capabilities of its models on well-specified tasks at around twice the rate it did between March 2024 and October 2025. The group is careful about the uncertainty. It says it would not be surprised if the true figure was closer to 1.3x or 3x the historical rate.
How the review was done
From August 12 to October 29, 2026, METR staff worked inside Anthropic on company devices, with access to internal benchmark scores, training and reinforcement-learning data, compute figures, forecasts for future models, and interviews with about 15 staff members. The team was small, about 2.5 full-time staff, and METR says it took no payment from Anthropic, which provided laptops and free tokens. The work was separate from the broader access arrangement Anthropic announced in October. METR also reviewed the automated research section of Anthropic's August 2026 risk report on the same day.
Why AI gets most of the credit
METR's best guess is that about 80 percent of the current acceleration comes from AI itself rather than from bigger computers or more staff. The group says it is confident the jump was not caused by faster growth in training compute. In other words, the speedup comes from using AI tools to do research work, not just from throwing more hardware at the problem. METR is clear about what it did not find. AI is not yet making fully autonomous contributions to the most important parts of model training. The acceleration is real, but it is AI-assisted research, not AI doing research entirely on its own. METR also saw no technical bottleneck, in algorithms, training compute, experiment compute, or data, that would stop progress over the next 12 months.
The redaction problem
There is a catch, and METR states it plainly. Anthropic had the right to redact any non-public information from the published report, and it used that right. METR says Anthropic redacted a majority of the evidence it found most informative for its conclusions, including per-stage acceleration estimates, the breakdown of compute, data and algorithms, bottleneck analyses, and forecasts for future models. METR says the public version does not meet its own transparency standard, so readers have to rely on its judgment more than usual. A planned check of how Anthropic allocates its compute was also cancelled after Anthropic changed the scope. The redactions do not make the assessment worthless, but they do limit independent scrutiny.
Why it matters
If AI keeps speeding up AI research, progress could arrive faster than timelines built on the old pace. That is why METR treats AI research acceleration as a key risk indicator, and why Anthropic lists it in its own safety policy. METR's review of Anthropic's risk report also notes it is plausible Anthropic had already crossed its 2x acceleration early-warning threshold, though many researchers involved lean toward the view that it had not. The report does not claim AI is now building itself, and it does not promise the speedup will continue forever. But it is the first independent, high-access look at how fast a leading lab is actually moving, and the answer is about twice as fast as before.
Sources
- METR - https://metr.org/blog/2026-11-05-anthropic-ai-rd-acceleration-assessment/ - TechBooky - https://www.techbooky.com/independent-review-finds-anthropic-ai-progress-may-be-twice-as-fast/ - The Neuron - https://www.theneuron.ai/digest/everything-that-happened-in-ai-today-friday-november-6-2026/