TL;DR
METR estimates that Anthropic’s model capability progress accelerated 1.9 times after October 2025 compared with its earlier rate. The nonprofit says AI assistance likely explains much of the change, but Anthropic redacted a majority of the evidence METR considered most informative.
METR estimated that Anthropic’s model capability progress accelerated 1.9 times after October 2025, compared with its prior pace, in an independent assessment published November 5. The nonprofit said AI assistance likely accounts for much of the increase, while warning that Anthropic redacted most of the evidence it found most informative, limiting what outsiders can verify.
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
METR’s estimate has a 95% confidence interval of 1.5 to 2.8 times. The group described its overall judgment as around twice the earlier rate and said it would not be surprised by a result ranging from 1.3 to 3 times. Its measure compares model capability progress over time; it does not mean researchers are twice as productive across every part of development.
The analysis used Anthropic’s internal benchmark scores for released and unreleased models, training metrics, surveys and about 15 staff interviews. METR said roughly 2.5 full-time-equivalent staff worked on the assessment over 11 weeks, from August 12 to October 29, using Anthropic devices. It said it took no payment; Anthropic provided laptops and free usage tokens and retained the right to redact non-public material.
METR’s measure of Anthropic’s benchmark performance rose from about 15 points per year to 29 points per year, with an estimated change point of October 7, 2025. For frontier models across developers, the estimated increase was smaller, from 17 to 24 points per year, or 1.4 times. METR also said Anthropic’s own August 2026 figures showed Claude leading 26% of its AI research and development work, up from less than 1% in February; the company reported about 30,000 agents active at a time on its main internal platform.
METR’s Anthropic Analysis Raises Questions About AI Progress
METR estimates Anthropic’s model capability progress accelerated after October 2025, with AI assistance likely contributing. Redactions limit how much of the evidence outsiders can verify.
What METR measured
The assessment used internal data, including scores from released and unreleased models.
METR examined benchmark scores, training metrics, surveys and staff interviews.
It placed the change point at October 7, 2025, and compared progress rates before and after.
METR said AI assistance likely explains much of the increase, but did not conclude that AI builds successor models autonomously.
A policy threshold under scrutiny
Anthropic’s Responsible Scaling Policy identifies twice-baseline progress with automation as a potential early warning.
- A threshold crossing by July 15, 2026 was plausible.
- That is METR’s interpretation, not a confirmed policy determination.
- A November 2 filing said it may imminently meet its definition.
- Anthropic says fully autonomous AI development of a successor has not been reached.
Limits on public verification
The findings matter for business and governance decisions, but the underlying evidence is incomplete in public.
Major redactions
METR said Anthropic redacted most of the evidence it considered most informative, limiting independent checks.
One measure of progress
Benchmark performance captures one aspect of development, not the full timeline or cost of AI research.
AI’s contribution is unclear
METR described AI’s share as perhaps about 80%, but the report does not show how that estimate varies under different assumptions.
The Stakes of Faster AI Research
The findings bear on whether AI tools are helping labs develop more capable AI systems at an accelerating pace. If that contribution grows, safety reviews and governance decisions may need to keep pace with shorter development cycles. METR’s estimate is an assessment of one company’s progress under a particular benchmark, not proof that AI systems can autonomously build their successors.
The distinction matters because Anthropic’s Responsible Scaling Policy sets an early-warning threshold for automated research and development. The policy describes a doubling of progress pace as one possible trigger, when the increase is plausibly substantially attributable to research or engineering automation. METR said it was plausible that Anthropic’s threshold had been crossed by July 15, 2026. That is METR’s interpretation of the available evidence, not a confirmed determination that the policy trigger was met.
Products Worth Considering
DJI Mini 5 Pro Fly More Combo with DJI RC 2
Kindle Paperwhite Signature Edition 32GB E-Reader, 7-Inch Display, Seaglass Green
Logitech MX Mechanical Wireless Illuminated Keyboard, Clicky Switches
As an affiliate, we earn on qualifying purchases.
Anthropic’s Two-Times Policy Threshold
Anthropic’s Responsible Scaling Policy version 3.4 took effect July 8, 2026. It identifies two automated R&D triggers: an AI system able to replace the company’s research staff at competitive cost, or progress advancing at twice the baseline pace with automation a substantial cause. The policy calls the threshold a potential early warning. Anthropic has said fully autonomous AI development of a successor has not been reached and is not inevitable.
METR’s work differs from typical outside evaluations, which test finished models. Its researchers examined internal company data, including results from models Anthropic had not released. METR’s Anthropic-only estimate was higher than the 1.7-times figure in the company’s Haiku 5.5 system card; METR said including newer, unreleased models was the main reason for the difference.
In a November 2 filing with the U.S. Securities and Exchange Commission, Anthropic said it “may imminently meet” its definition of dramatically accelerated progress. The filing and METR report indicate that both the company and outside evaluators are tracking acceleration, but they do not establish that the policy threshold has definitively been crossed.
“A 2x acceleration means that 2 years of historical progress is made in 1 year.”
— METR
Evidence Redactions Limit Verification
METR said Anthropic redacted a majority of the evidence it found most informative. The report therefore offers an estimate that outside readers cannot fully check against the underlying material. METR also dropped a planned analysis of how Anthropic divides computing resources between safety work and capability development.
The estimate of AI’s role is itself uncertain: METR said it was “confident” the acceleration did not come from faster growth in training compute, but described AI’s contribution as “perhaps” about 80%. The report does not establish how that share was calculated or how much it might vary under different assumptions. Its benchmark-based measure also captures one aspect of progress, not the full timeline or cost of AI research and development.
It remains unclear whether Anthropic considers its policy threshold crossed, what additional safeguards would follow if it did, and whether similar acceleration is occurring across other labs. The source material does not provide a final determination on these points.
Policy Decisions and Further Evidence
The next key development is whether Anthropic concludes that it has met the automated R&D threshold in its Responsible Scaling Policy and, if so, what safeguards it applies. The company’s November 2 filing said the threshold may be met imminently, while METR judged a crossing by July 15 plausible; neither statement amounts to a definitive public confirmation.
Further evidence could clarify how the trend continues, how much AI assistance contributes, and whether the pattern extends beyond Anthropic. Public scrutiny will also depend on whether more of the redacted evidence or additional independent evaluations become available. The timing and contents of any such disclosure remain unknown.
Key Questions
What did METR estimate?
METR estimated that Anthropic’s model capability progress accelerated 1.9 times after October 2025 compared with its earlier pace. Its 95% confidence interval was 1.5 to 2.8 times.
Did METR conclude that AI is building AI by itself?
No. METR said AI likely contributed substantially to faster research and development, but the report describes AI helping people build models. Anthropic says fully autonomous development of a successor has not been reached.
How much of the evidence can the public review?
METR said Anthropic redacted a majority of the evidence it found most informative. The report’s underlying data therefore cannot be fully checked by the public.
Has Anthropic confirmed that it crossed its policy threshold?
The source material does not report a definitive confirmation. Anthropic said in a November 2 filing that it “may imminently meet” its definition, and METR said a crossing by July 15 was plausible.
Source: Thorsten Meyer AI
