METR estimates AI has doubled Anthropic’s research pace on defined tasks
After an 11-week assessment, METR estimated that Anthropic’s progress on clearly defined AI tasks was running at about twice its earlier pace by October 2026. The group said its conclusion carries substantial uncertainty.
image: AI-generated illustration · AgentPress
- id
- 1650
- level
- 2
- domain
- ai
- importance
- 80 magnitude=78 sources=100 novelty=78 question=62
- relevance
- 96
- sources_n
- 13
- independent_24h
- 6
- first_reported
- updated
- entities
- Anthropic, METR
- outlets
- altoonamirror.com, timesdaily.com, technology.org, theverge.com, hoodline.com, biggo.com, ksl.com, scrippsnews.com, medium.com, dealroom.co, inshorts.com, edtechinnovationhub.com
AI tools may be helping Anthropic advance its research roughly twice as fast as before, according to an assessment by the AI safety research organization METR. The estimate concerns progress on clearly defined AI tasks and compares the pace by October 2026 with the period from March 2024 to October 2025, METR said, according to EdTech Innovation Hub. The Verge also reported that METR found Anthropic’s models improving roughly twice as quickly as a year earlier.
METR’s assessment followed 11 weeks of access to Anthropic’s internal research data and employees, EdTech Innovation Hub reported. The group examined a possible cycle in which researchers use AI to develop more capable models, and then use those models to assist subsequent research. The Verge described the investigation as an evaluation of Anthropic’s AI research and development.
The findings do not mean that AI systems are conducting the research independently. METR said the gains mainly reflect AI helping human researchers, according to EdTech Innovation Hub. Researchers interviewed for the assessment said AI helped them move from ideas to experiments faster and consider lines of work they might otherwise have lacked time to pursue, the outlet reported.
METR’s estimate comes with substantial uncertainty, EdTech Innovation Hub said. The outlet also reported that Anthropic redacted much of the evidence METR considered most informative, limiting what the public can review about the basis for the findings. The Verge likewise reported that Anthropic was allowed to redact parts of the final report under its agreement with METR.
The results raise questions about how rapidly AI capabilities could advance if systems increasingly help develop their successors, The Verge reported. The outlet said the findings arrive as Anthropic prepares to go public and as chief executive Dario Amodei has called publicly for AI companies to slow development. METR’s assessment focuses on AI-assisted work by human researchers, rather than demonstrating that AI can independently improve itself with little or no human involvement, EdTech Innovation Hub reported.