The evaluation nonprofit METR published an independent assessment of AI R&D acceleration at Anthropic on November 5. It concludes that Anthropic is likely accelerating its capabilities progress, that AI use is the majority driver of the current acceleration, and that there is a significant risk progress could speed up further because METR found no clear bottlenecks.
The assessment ran from August 12 to October 29, 2026, with METR staff working on Anthropic devices. Anthropic supplied internal benchmark data, scores on its version of the Epoch Capabilities Index (AECI), algorithmic efficiency data, compute and headcount metrics, uplift surveys and about 15 staff interviews. METR did not receive query access to unreleased models.
METR estimates that since October 2025, AECI growth for Anthropic models has been 1.9 times the earlier rate (95% confidence interval 1.5x to 2.8x), moving from about 15 to 29 points per year. Estimates ranged from 1.1x to 2.9x across analysis choices, and fell to 1.4x when all frontier developers' models were included. METR attributes about 80% of the October 2026 acceleration to AI and says it is confident the speedup did not come from faster growth in training compute. Self-reported uplift in internal polls rose from about 1.2x for Claude 3.7 Sonnet to about 12x for the latest internal model, which METR calls "very loose evidence."
METR is explicit about limits. Anthropic held unilateral redaction rights and redacted most of the evidence METR found most informative, so the public "has to rely on METR's judgment much more than would be ideal." Anthropic also cancelled a planned compute-allocation verification component midway through. METR took no payment but received free tokens and laptops, and disclosed a conflict of interest involving one team member.
The report is one of the first third-party attempts to measure whether a frontier lab's own models are speeding up its research. For developers and policymakers, the takeaway is that the cadence of capability gains from at least one major lab may continue to shorten, and that external verification of such claims still depends heavily on what labs choose to disclose.