METR: Anthropic AI R&D Pace Nearly Doubled

Independent evaluator METR said on November 5, 2026 that Anthropic is likely improving its AI models on well-specified tasks at about twice the rate it managed between March 2024 and October 2025, and that AI itself is probably the main reason. The finding comes from an 11-week, high-access review of Anthropic's internal data, though METR said Anthropic redacted most of the evidence it considered most informative.

Key takeaways

• METR estimates Anthropic's internal capability index has grown 1.9x faster since October 2025 (95% confidence interval: 1.5x to 2.8x).
• METR guesses "perhaps ≈80%" of the October 2026 acceleration is due to AI, and says faster growth in training compute does not explain it.
• Anthropic redacted most of the evidence METR found most informative and canceled a planned check of how it splits compute between capabilities and safety.

How fast is AI accelerating AI research at Anthropic?

METR's central measure is AECI, which the report describes as "Anthropic's version of the Epoch Capabilities Index," a composite score built from benchmark results. METR fit the index over time with a single breakpoint and compared the slopes before and after it.

The result: capabilities are "growing at an estimated 1.9x the pre-October 2025 rate [95% confidence interval: 1.5–2.8x]," according to the METR report. In plainer terms, METR wrote that it "would not be surprised if the current rate of growth was closer to 1.3x or 3x."

That estimate is higher than the 1.7x figure in Anthropic's own Haiku 5.5 system card. METR attributes the gap mainly to data from newer, unreleased models that it was allowed to see.

Is AI the cause, or just more compute?

METR said it was confident the speedup did not come from faster growth in training compute. Its best guess is that "a majority (perhaps ≈80%) of the current (October 2026) acceleration is due to AI," with substantial uncertainty. It added that the late-2025 shift was likely influenced by non-AI factors.

Staff interviews and internal self-reports were "consistent with" AI doubling the pace, the report said, though METR labeled self-reported uplift figures from internal polls as very loose evidence. It found no convincing bottleneck in algorithmic progress, compute or data for the next 12 months, but said it was not confident in that call.

What METR could see, and what the public cannot

The engagement ran from August 12 to October 29, 2026. METR received internal benchmark scores, headcount and compute metrics, code activity data and about 15 interviews with key staff, working on Anthropic devices. METR said it did not take payment from Anthropic; Anthropic supplied free tokens.

Anthropic held the right to redact non-public information. METR wrote that "Anthropic redacted a majority of the evidence we found most informative for these conclusions." Anthropic described the cuts as "aimed at removing confidential information that could accelerate other AI developers." METR conceded the public report therefore relies on its judgment more than would be ideal.

A separate component, verifying how Anthropic allocates compute between capabilities and safety work, was canceled by Anthropic before METR reached conclusions. Anthropic said it would revisit that later.

Context: Anthropic's evaluator program

Anthropic lists AI R&D acceleration as a key risk indicator under its Responsible Scaling Policy. In an October 15 update on independent evaluators, the company said it had onboarded six METR employees with "employee-like access." METR stressed that this assessment is a distinct project from that program.

The review follows Anthropic chief executive Dario Amodei's September call to pace the frontier after a July agent incident, and his pledge to let outside evaluators publish findings.

What's next

METR listed reasons it could be wrong in both directions. Its index may understate real-world impact, or the speedup may be limited to clean, well-specified tasks. The open questions are whether Anthropic restores the compute-allocation check and whether future reports disclose more of the underlying evidence.

FAQ

What does "2x" mean here?
It means Anthropic's internal capability score is rising about twice as fast per unit of time as it did from March 2024 to October 2025, not that models are twice as capable.

Did METR see Anthropic's unreleased models?
It saw benchmark data for internal-only models but did not get access to sample from non-public models.

Can readers check the evidence?
Only partly. METR says its roughly 100-page full report went to Anthropic, and most of the strongest evidence was redacted from the public version.

References

• Independent Assessment of AI R&D Acceleration at Anthropic in October 2026 — METR (Nov. 5, 2026)
• An update on independent evaluators — Anthropic (Oct. 15, 2026)
• Measurements for understanding the pace of AI development inside frontier labs — Anthropic