Report finds Anthropic is accelerating AI model development

The AI safety research organization METR released a report on Nov. 5, 2026, indicating that Anthropic is developing new AI models at a significantly faster pace than in previous years. The findings follow an 11-week investigation conducted between Aug. 12 and Oct. 29, 2026.

According to the report, the capabilities of Anthropic's models on specific tasks are improving at roughly twice the rate observed between March 2024 and October 2025. METR researchers calculated that benchmark scores have accelerated by a factor of 1.9x since October 2025, though they noted high uncertainty, stating the true rate could be between 1.3x and 3x the historical pace.

The investigation, which included access to internal model scores, code commits, and interviews with approximately 15 staff members, attributes about 80% of this acceleration to Anthropic’s use of AI to assist in its own research and development. METR noted that the company exercised redaction rights over a majority of the evidence found during the audit.

METR concluded that it is plausible Anthropic reached an early warning threshold by July 15, 2026, a claim that the company had previously downplayed in its own public reporting. Additionally, Anthropic noted in an amended IPO filing on Nov. 2, 2026, that it may soon meet the threshold for dramatically accelerated progress as defined by its internal Responsible Scaling Policy.

The report identified no immediate bottlenecks in compute, data, or algorithms that would likely halt this trend over the next 12 months. METR stated that a majority of the current acceleration is linked to the company's extensive use of AI to perform its own research.