How much does AI help the teams developing AI itself? METR’s November 5 assessment examines evidence from Anthropic’s internal research and development work. Its estimates suggest faster capability progress on specified tasks, with substantial uncertainty about the size and causes of the effect.
METR estimates an acceleration of roughly two times for the scoped work, with a broad uncertainty range. That is an assessment of a particular organization and period, not a measurement that every company or worker becomes twice as productive. AI assistance also does not mean the research process was fully autonomous.
Read the access and disclosure limits
The assessment describes the evidence available to METR and limits associated with confidential information. Anthropic could redact some information, and METR did not have access to every unreleased model. METR also discloses relationships and explains the financial arrangements for the assessment. Those details matter when weighing a difficult-to-reproduce analysis of internal work.
Our view: this is a valuable question, but broad claims should not outrun the evidence. Independent replication across organizations and transparent task definitions would be needed before treating these estimates as a general productivity rule. We have not independently audited the underlying private data.
Independent reporting based on the linked sources. Product and performance claims are attributed to their publishers unless noted.
