Anthropic speeds up AI development pace

The rate of improvement for Anthropic’s artificial intelligence systems has doubled in the past year, according to a new analysis from an independent research group. The finding comes as the San Francisco-based company prepares for a major stock market debut later this month.

Model Evaluation and Threat Research, a nonprofit based in Berkeley, California, released its findings on Thursday. The group was invited by Anthropic to review internal data and interview staff. This access allowed researchers to look closely at how fast the company is building new capabilities.

The timing is notable because Anthropic chief executive Dario Amodei has recently urged the industry to slow down. In an essay published on Sept. 12, Amodei argued that safety measures need more time to catch up with technological advances. He called on the AI world to reduce the pace at which model capabilities improve.

The research group estimated that Anthropic is improving the performance of its models on specific, well-defined tasks at twice the speed it was doing a year earlier. METR researchers believe that AI itself is helping to drive this faster development. Anthropic has stated that its Claude chatbot is increasingly used to help develop newer, more advanced versions of its own systems.

METR has evaluated models for Anthropic, OpenAI, and other labs for several years. The nonprofit is funded by donations and does not take payment for its work. However, partner companies do provide access to their systems and tokens, which METR uses for its testing and engineering efforts.

One part of the report was based on data provided by Anthropic over an 11-week period from mid-August to late October. The team also spoke with employees. The group said it received access to internal data on research and development that it had not had before. The assessment focused on two main questions: whether the growth rate of internal capabilities has accelerated, and how much of that change can be blamed on AI tools.

The report noted that Anthropic’s edits to the document limited how much detail METR could share. There were also some disagreements and uncertainty among the METR team. Despite these factors, the group concluded that the company’s rate of improvement had doubled. The report stated that Anthropic made as much progress on benchmarks in the last year as it did between October 2023 and October 2025.

The researchers ruled out increases in computing power as the cause of this acceleration. They found no bottlenecks such as limits on data or processing power holding back progress. A key concern is that using AI to improve AI could start a compounding effect that speeds up development further. The report said AI is not yet making fully autonomous contributions to the most critical parts of training, but Anthropic researchers described signs that this capability may arrive soon.

Amodei addressed this issue in his September essay by defining recursive self-improvement as AI’s growing ability to build the next generation of AI. He warned that if left unchecked, this process could outrun human ability to understand and control the systems. He also supported the idea of having external evaluators like METR embedded within leading AI labs to provide an independent check.

OpenAI chief Sam Altman said at the time that he also supported independent evaluators with employee-like access. Anthropic did not immediately respond to messages seeking comment on the new report. The company is scheduled to go public later this month in what could be the largest initial public offering in history.