METR estimates Anthropic's rate of capability progress has roughly doubled

METR, the AI-evaluation nonprofit, published two documents on 5 November from a review it ran at Anthropic between 12 August and 29 October. The first is a holistic assessment of how far AI is speeding up AI research there. It concludes that Anthropic is “likely improving the capabilities of their models on well-specified tasks at around twice the rate they did between March 2024 and October 2025”, with high uncertainty: METR said it would not be surprised by 1.3x or 3x. The second reviews the automated-R&D section of Anthropic’s August risk report.

The context was Anthropic’s amended stock-market filing three days earlier. Amendment No. 1 to its S-1, part of its public IPO filing, stated:

our internal estimates and our expectations regarding future models, rather than our current models, indicate that we may imminently meet our Responsible Scaling Policy’s definition of “dramatically accelerated progress attributable to AI.” We have not designated any of our current models as exceeding this threshold, and we continue to assess the underlying data.

— Anthropic, Form S-1/A, risk factors

METR’s main quantitative finding is that scores on AECI, Anthropic’s version of the Epoch Capabilities Index, have grown since October 2025 at about 1.9 times the earlier rate (95% confidence interval 1.5 to 2.8x). It guessed that a majority, perhaps about 80%, of the current acceleration is due to AI, and said it was confident the cause was not faster scaling of training compute. It also warned that the benchmark trend could be an artefact of the index or of overfitting, and that it was unsure how far the speed-up carries over to less well-specified work. In the second review METR agreed that catastrophic risk from models available on 15 July was low. It said most relevant researchers at METR and Anthropic leaned towards the report’s 2x early-warning threshold not having been crossed by then, but that METR thought it plausible it had been, and that the public report “understated the evidence for acceleration”.

The access had limits, which METR set out itself. Anthropic held unilateral redaction rights, and METR said Anthropic “redacted a majority of the evidence we found most informative”; the public “has to rely on METR’s judgment much more than would be ideal”. Anthropic described the redactions as aimed at confidential information that could accelerate other developers. METR did not get query access to unreleased models, and said this engagement was distinct from the arrangement Anthropic announced on 15 October. It also disclosed that one team member had discussed a role at Anthropic, and that their work was checked by a colleague.

The outside estimate gave a number to a claim the company had so far made only in qualified terms, and it arrived after METR’s earlier review of the misalignment section of the same report.