🔍 Read the full analysis: Is AI Starting To Speed Up AI Research? METR Examines Anthropic on ThorstenMeyerAI.com
Get business pricing on monitors, keyboards and dev gear
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
METR’s November 5 assessment estimates that Anthropic’s capability progress on well-specified tasks accelerated to about 1.9 times its March 2024–October 2025 rate. METR believes AI use likely accounts for much of the increase, but says Anthropic redacted a majority of its most informative evidence, leaving key details difficult for outsiders to verify.
METR, an independent AI evaluation nonprofit, estimates that Anthropic’s progress on well-specified AI capability tasks has accelerated to about 1.9 times its previous pace, based on an 11-week assessment using internal company data. The report, published November 5, says AI assistance likely accounts for most of the speed-up, but also says Anthropic redacted a majority of the evidence METR found most informative, limiting what outsiders can independently assess.
METR’s main estimate compares Anthropic’s capability trend before and after a change point it places at October 7, 2025. Using an index derived from benchmark scores, the report estimates progress at about 15 points per year before that date and roughly 29 points per year afterward. METR describes the result as 1.9x acceleration, with a 95% confidence interval of 1.5x to 2.8x. It says it would not be surprised by estimates ranging from 1.3x to 3x.
The result is not a direct measure of how quickly Anthropic’s researchers work, nor does it show that AI systems are independently designing their successors. METR says the estimate concerns progress on well-specified tasks. The report’s interpretation is that perhaps 80% of the acceleration is due to AI use; METR presents that as an estimate, not a directly measured share. It says it is confident the change did not result from faster growth in training compute.
METR says its team used benchmark results for unreleased models, training metrics, surveys and about 15 staff interviews. About 2.5 full-time-equivalent staff worked on the assessment from August 12 to October 29, using Anthropic-provided devices. METR says it received no payment. Anthropic supplied laptops and free model tokens and retained the right to redact non-public information. The report says the company removed a majority of the evidence METR considered most informative and that a planned analysis of the division between safety and capability compute was dropped.
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 Faster Model Progress Could Mean
If the estimate is broadly right, AI tools are helping accelerate some of the work that produces more capable AI systems. That matters because faster capability development could shorten the time available for evaluation, safety research and policy responses. METR’s report does not establish that development is fully autonomous or that every part of AI research is speeding up at the same rate.
The finding also bears on Anthropic’s own safety framework. The company’s Responsible Scaling Policy identifies a doubling in the pace of progress, plausibly attributable to automated research or engineering, as one trigger related to automated research and development. METR says it is plausible that this early-warning threshold had been crossed by July 15, 2026. That is METR’s assessment, not a confirmed determination by Anthropic that its policy trigger has been met.
The evidence limits are part of the story. METR had access to internal material, but the public version omits much of what its evaluators considered most useful. Readers therefore have an external assessment based on unusually direct access, alongside substantial restrictions on public verification.
Popular right now
Breville Bambino Espresso Machine, Brushed Stainless Steel
Breville BES870XL Barista Express Espresso Machine
Dyson Big Ball Multi Floor Canister Vacuum, Yellow
As an affiliate, we earn on qualifying purchases.
How METR Measured Anthropic’s Pace
METR’s approach differs from an evaluation of a publicly released model. Its staff worked with internal company information, including performance data for models that had not been released. The assessment combined those records with interviews and surveys, then compared the company’s measured capability trend across time. The report’s index is based on benchmark performance; it is not a count of research hours saved or experiments completed.
Anthropic’s Responsible Scaling Policy, version 3.4, took effect July 8, 2026. As described in the source material, it identifies two conditions related to automated research and development: a model capable of replacing the company’s research staff at competitive cost, or a doubling of the pace of progress plausibly attributable in substantial part to research or engineering automation. Anthropic’s risk report describes the threshold as a potential early warning, rather than proof that the associated risk is already present.
Anthropic separately reported that Claude “leads” 26% of its AI research and development work in August 2026, compared with under 1% in February. The company also reported about 30,000 agents operating at a time on its main internal platform. Those are Anthropic’s own measurements, and the source material does not define precisely what “leads” means or how the figures were calculated.
““We would not be surprised” by an acceleration estimate between 1.3x and 3x.”
— METR
The Evidence Behind 1.9x
The estimate remains sensitive to measurement choices. METR reports that changing assumptions about task difficulty moves the acceleration estimate from 1.1x to 2.9x, although it says most other checks remain near 2x. When it considers frontier models across developers rather than Anthropic alone, the estimated rate changes from 17 to 24 index points per year, or about 1.4x. METR says this could reflect Anthropic catching up from a weaker position in 2025, or earlier Anthropic models being less tuned to benchmarks; it does not resolve which explanation is correct.
Much of the supporting evidence is not public, and METR dropped its planned examination of how Anthropic divides compute between safety and capabilities. The source material also does not provide enough detail to independently evaluate the company’s “leads” metric for Claude or the reported agent count. It is not clear whether Anthropic has formally concluded that its Responsible Scaling Policy’s doubling trigger has been met. The findings concern Anthropic and should not be treated as a measurement of the entire AI industry.
Anthropic’s Policy Threshold
The next key development is whether Anthropic formally determines that its automated R&D criteria have been met and what measures it would take under its Responsible Scaling Policy. The company’s November 2 filing said it may imminently meet its definition of dramatically accelerated progress, while METR’s report says the threshold may plausibly have been crossed by July 15. Neither statement, as described in the source material, amounts to a public confirmation that the trigger has been formally declared.
Further scrutiny will depend on whether Anthropic releases additional evidence or clarifies the measures behind its internal figures. METR’s public report provides its conclusions and methods, but the redactions and omitted compute analysis leave outside researchers with limited ability to check key parts of the assessment. No further release or review date is specified in the source material.
Key Questions
What does METR’s 1.9x estimate mean?
METR estimates that Anthropic’s progress on its measured capability index rose from about 15 points per year before October 7, 2025, to around 29 points per year afterward. The report describes that as roughly 1.9 times the earlier pace, with a 95% confidence interval of 1.5x to 2.8x.
Does the report show that AI is building the next AI on its own?
No. METR’s assessment concerns progress on well-specified tasks and its interpretation that AI likely contributed substantially to the acceleration. The source material describes current development as AI assisting human researchers, not fully autonomous design of a successor system.
How much of the acceleration does METR attribute to AI?
METR estimates that perhaps about 80% is due to AI, while saying it is confident faster growth in training compute did not cause the change. The 80% figure is an attributed estimate, not a definitive causal measurement.
Has Anthropic confirmed that its safety-policy threshold was crossed?
The source material does not establish that Anthropic has formally declared the threshold met. METR says it is plausible the threshold was crossed by July 15, while Anthropic said in a November 2 filing that it may imminently meet its definition of dramatically accelerated progress.
Why are the findings hard to verify independently?
METR says Anthropic redacted a majority of the evidence it found most informative. A planned analysis of how the company allocates compute between safety and capabilities was also dropped, leaving important supporting details unavailable to the public.
Source: ThorstenMeyerAI.com
