metr anthropic acceleration: AI Is Doubling Anthropic's R&D Pace

The metr anthropic acceleration report is the AI story everyone in safety circles is arguing about this weekend. METR, an independent AI evaluation group, spent 11 weeks inside Anthropic and concluded the lab is likely improving its models at around twice the historical rate, mostly because Anthropic uses its own AI to do research. The catch: Anthropic redacted most of the evidence METR found most informative.

What actually happened

METR published "Independent Assessment of AI R&D Acceleration at Anthropic in October 2026" on November 5. Its staff ran a high-access evaluation from August 12 to October 29, with access to internal metrics on pretraining, reinforcement learning, compute, benchmark scores for internal models and staff interviews.

The headline finding: benchmark scores on AECI, Anthropic's version of the Epoch Capabilities Index, have grown at about 1.9x the pre-October 2025 rate since then, with a 95% confidence interval of 1.5x to 2.8x. METR says that implies roughly two years of historical progress in one year, and that it is confident the jump was not driven by faster scaling of training compute.

The Verge's November 7 explainer added color. Employees using Anthropic's most capable internal model self-reported being around 12 times as productive on average, roughly double what they reported a few months earlier. METR also said Anthropic has finished training three unreleased models in the past six months, codenamed Model 2, Model 3 and Model 4.

The redaction fight

METR's own redaction summary is blunt: Anthropic had the right to redact non-public information, and it "redacted a majority of the evidence we found most informative." Anthropic described the redactions as removing confidential details that could accelerate other developers.

Anthropic cofounder Sam McCandlish responded on X, as quoted by The Verge, that sharing exactly what drives the speedup "risks handing out a roadmap for others to go faster." He said Anthropic plans to publish its own report on the speed of AI progress in the coming weeks. Critics were not impressed. Daniel Kokotajlo, who runs the AI Futures Project behind the AI 2027 scenario, accused Anthropic of "recklessly gambling with our lives," per The Verge.

We covered METR's earlier arrangement with Anthropic in anthropic metr independent evaluators; The Verge notes this probe is separate from that "employee-like" access program. The report lands while Anthropic prepares to go public, as tracked in our anthropic ipo date odds piece.

What Polymarket thinks

There is a live market on the scariest version of this story. At our ~06:04 UTC refresh on November 8:

• "Will OpenAI or Anthropic announce recursive self-improvement by December 31, 2026?" traded near 8% Yes.
• A separate "OpenAI vs. Anthropic: first to announce recursive self-improvement?" market sat around 54% for OpenAI.

Both books are tiny, with only a few thousand dollars in volume and no trades in the last 24 hours at our snapshot. Treat them as vibes with a price tag, not a consensus.

METR did not say recursive self-improvement is here. It said AI is already speeding up AI research, and it did not see obvious bottlenecks stopping that from getting faster.

Why the metr anthropic acceleration findings matter

For builders, the takeaway is pace. If a frontier lab is compressing two years of progress into one, model releases, price cuts and agent features will keep arriving faster than most product roadmaps assume.

For policy, it is ammunition. The Verge reported that New York Assemblyman Alex Bores called on Congress to "ban RSI," and the incoming Congress includes many members who campaigned against AI companies. Expect this report to show up in hearings.

For markets, it is a reminder that AI tokens and AI equities trade on narratives like this one. The metr anthropic acceleration headline can move sentiment even if nothing ships this week.

What to do as a reader

Read the METR report itself, including the redaction summary, before trusting any thread that claims to know what was hidden. Then compare it with The Verge's reporting and with Anthropic's own report once it appears.

If you trade the related Polymarket contracts, remember they are illiquid and resolve on a public announcement, not on what METR estimates. Nothing here is investment advice.