Summary
Matthew Berman walks through a new independent report from METR assessing AI R&D acceleration at Anthropic as of October 2026. The report is among the first results of third-party evaluators getting deep internal access to a frontier lab, though Anthropic could approve access and decide what was published.
METR staff conclude that Anthropic is likely improving its models on well-specified tasks at around twice the rate seen between March 2024 and October 2025, with plausible estimates ranging from 1.3x to 3x. The report estimates roughly 80% of the current acceleration is attributable to AI itself rather than extra hiring, compute or human research breakthroughs. The video points to an inflection around Claude Opus 4.5 in late 2025, with progress continuing through the Mythos release earlier this year, as shown on Anthropic’s internal capability index.
The video also covers evidence that the acceleration is not explained by training compute scale, though Anthropic redacted details, plus employee interviews. METR reports finding no clear bottlenecks or steep diminishing returns that would stop compounding progress, and warns of a significant risk of even faster acceleration in the coming year. Berman discusses what this could mean for AI over the next one to two years.
📺 Source: Matthew Berman · Published November 07, 2026
🏷️ Format: News Analysis
