Independent Assessment of AI R&D Acceleration at Anthropic in October 2026

METR conducted a high-access assessment at Anthropic from August to October 2026 to determine whether AI is speeding up AI research and development, a change that could make dangerous capabilities emerge more quickly. Its central judgment is that Anthropic is improving models on well-defined tasks at roughly twice the pace seen from March 2024 to October 2025, with a plausible range of 1.3 to 3 times. METR attributes most of the current acceleration - perhaps 80% - to AI assistance, while judging that non-AI factors likely contributed substantially to the initial acceleration in late 2025. Training-compute growth did not explain the change. Staff interviews and surveys point to major productivity gains from AI tools, primarily as assistants rather than autonomous researchers, though autonomous contributions may grow.

Anthropic’s benchmark index showed progress since October 2025 at 1.9 times its earlier rate, equivalent to about two years of historical progress in one year. METR cautions that benchmark design, sparse and lagging measurements, and the focus on verifiable tasks limit what this says about broader real-world capabilities. It found no clear near-term barriers in algorithms, compute, or data that would prevent further acceleration, raising the possibility of a compounding cycle in which stronger models speed up their own development. However, many of the most informative findings were redacted at Anthropic’s request, leaving the public unable to inspect key evidence; METR also identifies possible non-AI explanations and weaknesses in its analysis.