From mid-August to late October 2026, the evaluation organization METR carried out a high-access assessment inside Anthropic, maker of the Claude models, to measure how much artificial intelligence is accelerating AI research and development, meaning the work of designing and training new models. Anthropic shared internal benchmark scores, training compute measurements (the computing power used to build each model), productivity data, and staff interviews. The study follows questions METR proposed in October for judging whether AI is accelerating its own development.
METR's central conclusion is that Anthropic is now improving its models on well-specified tasks at roughly twice the rate of March 2024 to October 2025, with high uncertainty: the true acceleration could plausibly be 1.3x or 3x. A key evidence source is AECI, Anthropic's version of the Epoch Capabilities Index, a composite of benchmark scores. Since October 2025, AECI has grown at an estimated 1.9x its earlier rate (95 percent confidence interval 1.5 to 2.8), meaning Anthropic compressed about two years of earlier progress into one. METR is confident the speedup was not caused by faster growth in training compute.
Attributing the speedup to AI itself
METR guesses that a majority, perhaps around 80 percent, of the current acceleration comes from Anthropic's extensive use of AI in its own research, while non-AI factors likely played a large role in late 2025. Interviews and surveys are consistent with AI doubling the pace of progress, and staff self-reports of speedup, a loose measure, reached 12x in October 2026. Researchers said AI mostly assists humans rather than working fully autonomously, but described signs that autonomous contributions could arrive soon. METR found no clear bottleneck, such as limits on data, algorithms, or computing, that must slow progress within twelve months, though it is not confident in this judgment.
Anthropic exercised its right to redact a majority of the evidence METR found most informative, describing the redactions as aimed at removing confidential information that could accelerate other AI developers. METR disclosed this under its transparency approach and cautioned that the public should rely on its judgment. This matters because AI that accelerates AI research could compound into far faster progress than society expects, and independent assessments like this one may be the public's main early warning of such a shift.