METR says AI is roughly doubling Anthropic's model development speed
METR researchers with internal access assessed that AI accounts for about 80% of Anthropic's recent acceleration in model development.
METR assessed that AI accelerates Anthropic's model development by about 2x.
From August to October, several METR staff worked inside Anthropic with expanded access to internal data. Their task was to assess how much the growth of internal model capabilities had accelerated and what share of that acceleration is attributable to AI.
Key results:
Anthropic's pace of model improvement by October 2026 doubled compared with the period from March 2024 to October 2025. The estimate is approximate; a range of 1.3x–3x is cited. The main metric confirming this is AECI, an internal version of Epoch AI's capability index. Since October 2025, it has been growing 1.9x faster than before. In other words, over the past year Anthropic advanced roughly as much as it had over the previous two years.
At the same time, about 80% of the current acceleration comes from using AI in research and development. Staff themselves estimate, on average, that AI speeds up their work by about 2x. In Slack polls, individual estimates rose from 1.2x with Claude 3.7 Sonnet to 12x with the newest internal model (12x was equated to “I can do in a day what used to take almost two weeks”).
METR believes the acceleration can continue. They did not find obvious limits to this, at least for the next year. No clear ceiling is visible in algorithms, compute, or data.
Why it matters
AnalysisIf AI is already doubling R&D speed inside a leading lab, the pace of capability gains, competitive pressure, and safety risks may accelerate faster than external observers expect.
Discuss in community
Share your questions and insights with developers
Related in Ecosystem
Practical guides, core terminology, and agent skills
Claude Haiku (Fast Compact Models by Anthropic)
The most compact and fastest model in the Anthropic lineup, designed for lightning-fast responses, mass text classification, data extraction, and agent routing at minimal cost.
Claude Opus (Flagship AI Model by Anthropic)
The most powerful heavyweight model in the Anthropic family, designed for complex philosophical and scientific analysis, multi-layered text generation, and deep contextual understanding.
Avoiding KYC in Claude Ultra: VPS proxy and Anthropic domain blocking
A community-described method claims to bypass KYC in Claude Ultra by routing all traffic through a single VPS, blocking Anthropic domains and disabling telemetry.
Community Discusses Anthropic Bans Over the Past Week
Comments mention users reporting accounts banned by Anthropic during the past week, with no details provided.