🔍 Read the full analysis: How Much Is AI Speeding Up AI Research? METR’s Anthropic Study on ThorstenMeyerAI.com
Get hardware and tech essentials delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
METR’s independent assessment estimates that Anthropic’s capability progress on well-specified tasks accelerated to 1.9 times its earlier pace, with a 95% confidence interval of 1.5x to 2.8x. METR says AI use likely accounts for much of the change, but Anthropic redacted most of the evidence the evaluators considered most informative, limiting public scrutiny.
METR, a nonprofit that evaluates frontier AI systems, published an independent assessment on November 5 estimating that Anthropic’s AI capabilities are improving about 1.9 times faster than they did from March 2024 to October 2025. METR says AI use likely accounts for much of the acceleration, but Anthropic redacted a majority of the evidence the evaluators found most informative, leaving the public unable to check many of the underlying details.
METR’s central estimate compares the rate of progress before and after a change point it dates to October 7, 2025. Using Anthropic’s capability index, the assessment estimates a rise from about 15 points of progress per year before that date to about 29 points per year afterward. METR describes the result as roughly two years of historical progress in one year. Its estimated acceleration is 1.9x, with a 95% confidence interval of 1.5x to 2.8x; the organization says it would not be surprised by estimates ranging from 1.3x to 3x.
The work drew on internal benchmark scores for unreleased models, training metrics, staff surveys and roughly 15 interviews. METR says about 2.5 full-time-equivalent staff worked on the assessment over 11 weeks, using Anthropic devices. The nonprofit says it took no payment; Anthropic supplied laptops and free model tokens and retained the right to redact non-public information.
METR estimates that perhaps about 80% of the acceleration is attributable to AI use, while saying it is confident the change did not result from faster growth in training compute. That proportion is an estimate, not a directly measured share. Anthropic’s own figures, as reported in the source material, say Claude led 26% of its AI research and development work in August 2026, compared with less than 1% in February. The company reported about 30,000 agents working at a time on its main internal platform.
Infographic 01 · How fast is AI speeding up AI?
About twice the old pace, most likely because of AI
An outside evaluator's estimate from internal Anthropic data, plus Anthropic's own measurements. Wide uncertainty, much of the evidence redacted.
Acceleration of Anthropic’s benchmark progress (AECI) since October 2025, 95% confidence interval 1.5x to 2.8x. METR’s overall judgment: around twice the old rate; it “would not be surprised” by 1.3x or 3x.
Capability points per year
AECI trend of Anthropic models, break point 7 Oct 2025
All labs before October 2025: +17 a year, so 1.4x. Different task-difficulty assumptions: 1.1x to 2.9x.
Sources: METR assessment, 5 Nov 2026 · METR risk report review · Anthropic Institute measurements.
Powered by Thorsten Meyer AI
What the Acceleration Estimate Signals
The assessment bears on whether AI systems are helping researchers build more capable systems quickly enough to change the pace of AI development. If AI-assisted research contributes substantially to faster progress, the same tools could support a feedback loop: more capable models help with research and engineering that produce later models. METR’s findings suggest that this process may already be affecting the pace at Anthropic, but do not establish that AI systems are independently developing their successors.
The estimate also matters because Anthropic has a policy trigger tied to a doubling in progress that is plausibly attributable to automated research or engineering. METR considers it plausible that the company’s early-warning threshold had been crossed by July 15. Anthropic said in a November 2 filing that it “may imminently meet” its definition of dramatically accelerated progress. The assessment does not settle whether the policy’s formal conditions have been met, and the company’s wording remains a possibility rather than confirmation.
For readers and policymakers, the report shows both the value and limits of outside evaluation using company data. METR was able to examine information unavailable in public model releases, including unreleased-model results. But the redactions mean outsiders cannot independently verify much of the evidence behind the report’s most consequential conclusions.
Products Worth Considering
Apple MacBook Pro 2021 14-inch with M1 Max Chip, 64 GB RAM, 2 TB SSD, Silver
Apple MacBook Air 13-inch Laptop with M5 Chip, 16GB Memory, 512GB SSD, 12MP Center Stage Camera, Touch ID, Wi-Fi 7, Midnight
Apple MacBook Pro 16-inch with M1 Pro, 10-core CPU, 16-core GPU, 16GB RAM, 1TB SSD, Silver
As an affiliate, we earn on qualifying purchases.
How METR Measured Anthropic’s Progress
METR’s primary measure, AECI, combines results from multiple benchmarks into a single capability index. The report compares the index’s trend before and after its estimated October 2025 change point. This is a measure of progress on well-specified tasks, not a direct measure of researcher productivity across every part of AI development. METR defines a 2x acceleration as making in one year the progress that would historically have taken two.
The headline result is not the only estimate in the report. When METR examined frontier models across developers rather than Anthropic models alone, it found a smaller change: from about 17 to 24 index points per year, or 1.4x. METR says this may reflect Anthropic catching up from a weaker position in 2025, or differences in how earlier models were tuned for benchmarks. Under different assumptions about task difficulty, its estimate for Anthropic’s acceleration ranged from 1.1x to 2.9x.
Anthropic’s Responsible Scaling Policy, version 3.4, took effect July 8, 2026, according to the source material. It identifies two possible automated-R&D triggers: an AI system capable of replacing the company’s research staff at competitive cost, or a doubling in the pace of progress plausibly attributable to automated research or engineering. The policy describes the threshold as an early warning; it is not itself proof that a specific danger has arrived. Anthropic says fully autonomous development of a successor model has not been reached and is not inevitable.
““A 2x acceleration means that 2 years of historical progress is made in 1 year.””
— METR
Redactions Limit Public Verification
The main limitation is access to evidence. METR says Anthropic redacted a majority of the material it found most informative. The source material also says the company dropped a planned check of how it divides compute between safety work and capability development. The report’s public version therefore does not allow readers to inspect all the data or methods that informed its conclusions.
The 1.9x figure depends on the benchmarks, comparison period and assumptions used. METR’s lower 1.4x estimate across all developers and its 1.1x-to-2.9x range under different task-difficulty assumptions show that the result is not a single, assumption-free measurement. The report also does not establish that every part of Anthropic’s research process is faster, or that AI tools alone caused the change.
It remains unclear from the available material whether Anthropic’s policy threshold has formally been triggered, what additional evidence might change METR’s estimate, and how much of the redacted information could be released without exposing non-public company data. The assessment concerns Anthropic and should not be treated as a measurement of every AI lab.
Policy Review and Further Evidence
The immediate next step is determining whether Anthropic considers its Responsible Scaling Policy’s automated-R&D criteria met and what safeguards, if any, follow. The company’s November 2 filing says the threshold may be imminent; METR’s report says crossing it by July 15 is plausible. Neither statement, on the information provided, confirms that Anthropic has formally declared the trigger reached.
Further public evidence could clarify the estimate, including additional benchmark results, methods for attributing progress to AI assistance and information about how research resources are allocated. METR’s assessment is a dated snapshot based on work completed from August 12 through October 29, 2026. Whether the pace continues, changes or proves specific to Anthropic remains unestablished.
Key Questions
What did METR estimate?
METR estimated that Anthropic’s capability progress on well-specified tasks accelerated 1.9x relative to its earlier pace, with a 95% confidence interval of 1.5x to 2.8x. It estimates progress of about 29 index points per year after its October 2025 change point, compared with about 15 points per year before.
Does the report prove that AI is building AI by itself?
No. METR says AI likely contributes substantially to faster research and development, but the report describes AI assisting human-led work. Anthropic says fully autonomous design and development of a successor has not been reached and is not inevitable.
How much of the acceleration does METR attribute to AI?
METR’s estimate is “perhaps ≈80%.” That is an attributed estimate, not a directly measured share. The nonprofit also says it is confident the acceleration did not come from faster growth in training compute.
Why is the finding difficult to verify independently?
METR says Anthropic redacted a majority of the evidence it found most informative. The assessment used internal data, including results for unreleased models, that are not available for public review.
Has Anthropic confirmed that its policy threshold was crossed?
Not in the information provided. METR says a crossing by July 15 is plausible, and Anthropic said on November 2 that it “may imminently meet” its definition. Those statements do not confirm a formal trigger declaration.
Source: ThorstenMeyerAI.com
