An 11-week audit by nonprofit research group METR estimates that Anthropic’s model-development pace has accelerated by roughly 1.9 times since October 2025. The estimate would place the company close to a warning threshold in its own Responsible Scaling Policy for AI-driven research acceleration.
METR analyzed internal benchmark results, training metrics, surveys and interviews, including data from three unreleased models. It attributes about 80% of the measured acceleration to researchers using AI tools, rather than to fully autonomous research. The auditors found no clear bottleneck likely to stop further gains over the next year.
The result carries substantial uncertainty. METR says the true factor could be between 1.3 and 3, while an industry-wide estimate that includes competitors falls to about 1.4. Some Anthropic researchers also disputed colleagues’ self-reported productivity gains and credited hiring or other factors.
Public scrutiny is further limited because Anthropic redacted much of the evidence METR considered most informative. The audit therefore signals a possible acceleration, not a conclusive measurement that outsiders can independently reproduce.