TL;DR
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 estimates that Anthropic’s progress on measured AI capabilities has accelerated to about 1.9 times its earlier pace, with much of the change likely linked to AI-assisted research and development. The assessment draws on 11 weeks of internal data, but Anthropic redacted much of the evidence METR considered most informative, limiting public scrutiny.
METR, a nonprofit that evaluates frontier AI systems, estimated on November 5 that Anthropic’s measured capability progress accelerated about 1.9 times after October 2025. The assessment, based on 11 weeks of access to internal company data, attributes much of the estimated increase to AI helping with AI research and development, while noting that Anthropic redacted much of the evidence METR found most informative.
METR compared Anthropic’s rate of progress on its version of the Epoch Capabilities Index, which combines results across benchmarks. It estimates the rate rose from about 15 index points per year before October 2025 to about 29 points per year afterward. METR’s estimate for the acceleration is 1.9x, with a 95% confidence interval of 1.5x to 2.8x. The group said it would not be surprised by a result between 1.3x and 3x.
The estimate is higher than the 1.7x figure in Anthropic’s Haiku 5.5 system card, which METR says reflects its inclusion of newer, unreleased models. Looking across frontier models from all developers, rather than Anthropic alone, METR estimated a smaller increase: from 17 to 24 index points per year, or about 1.4x.
METR says it reviewed benchmark results for unreleased models, training metrics, surveys and about 15 interviews with Anthropic staff. Around 2.5 full-time-equivalent staff worked on the assessment using Anthropic devices. METR said it received no payment; Anthropic supplied laptops and free tokens and retained the right to redact non-public information.
Inside Anthropic METR’s Pace of Progress
METR estimates that Anthropic’s measured AI capabilities accelerated after October 2025. Redactions limit public review of the evidence behind that estimate.
What the estimate means
METR compared capability progress rates on its index before and after October 2025.
- About 15 index points per year
- Baseline before October 2025
- About 29 index points per year
- Estimated 1.9× increase
AI joins the research loop
METR estimates perhaps 80% of the acceleration is attributable to AI use, but this is not a directly measured causal share.
METR links much of the estimated increase to AI-assisted research and development.
Anthropic says current systems assist people building later systems; they do not autonomously develop their own successors.
Signals from Anthropic
Company-reported figures show AI taking a larger role in its research and development.
Claude leads 26%
Anthropic said Claude led this share of its AI research and development work in August 2026, up from under 1% in February.
About 30,000 agents
Anthropic reported this many agents working at any one time. The figure alone does not establish their impact on capability gains.
Evidence has limits
The estimate is an outside assessment based on unusual access to internal company data.
| What METR reviewed | What remains uncertain |
|---|---|
| Benchmark results, training metrics, surveys and staff interviews | Much of the most informative evidence was redacted |
| Anthropic’s measured capability trend | How much specific AI tools caused the change |
| Progress across frontier developers: 17 to 24 points per year | How the pace will evolve and whether a policy threshold was formally triggered |
AI Is Joining the Research Loop
The assessment concerns whether AI tools are helping researchers build more capable AI systems faster. METR estimates that perhaps 80% of the acceleration it observed is attributable to AI use. That is the organization’s estimate, not a directly measured share, and METR says faster growth in training compute did not account for the change.
Anthropic’s own reported usage offers another indicator of how AI is entering its development process. The company said Claude “leads” 26% of its AI research and development work in August 2026, up from under 1% in February. Anthropic also reported about 30,000 agents working at any one time on its main internal platform. Those figures describe the company’s measurements; they do not by themselves establish how much the agents caused the capability gains.
Faster progress could affect how quickly companies need to evaluate systems and apply safeguards. Anthropic’s Responsible Scaling Policy includes an early-warning threshold for progress doubling when the increase is plausibly tied substantially to automated research or engineering. METR’s estimate makes that policy relevant to current debate, though the assessment does not establish that any particular risk has materialized.
Products Worth Considering
Sonos Era 100 SL Compact Microphone-Free Speaker, White
EcoFlow DELTA 3 Portable Power Station (1024Wh, 1800W)
Apple iPad mini (A17 Pro), 128GB, Wi-Fi 6E, Space Gray
As an affiliate, we earn on qualifying purchases.
How Anthropic Defines Its Threshold
Anthropic’s Responsible Scaling Policy version 3.4, effective July 8, 2026, identifies two possible triggers related to automated research and development. One is a model capable of replacing the company’s entire research staff at competitive cost. The other is a doubling of the pace of progress, defined as achieving in one year what would otherwise take two years at the baseline rate, with the acceleration plausibly substantially attributable to automated research or engineering.
Anthropic describes recursive self-improvement as an AI system autonomously designing and developing its own successor. It has said that current systems remain short of that point: AI assists people building later systems, while people still choose the goals. The company has also said recursive self-improvement is not inevitable. Its risk report calls the doubling threshold a potential early warning, rather than proof that a danger is already present.
METR’s work differs from the more common external evaluation of released models because it used internal company data. The assessment estimated a break point around October 7, 2025, after which the measured trend rose. METR said Anthropic’s redactions prevented it from sharing most of the evidence it considered most informative, and it dropped a planned examination of how the company divides compute between safety and capability work.
“A 2x acceleration means that 2 years of historical progress is made in 1 year.”
— METR
Redactions Limit Public Review
The public cannot inspect much of the underlying material: Anthropic redacted a majority of the evidence METR found most informative. That restricts outside readers’ ability to assess the data and methods behind the estimate. METR also did not complete its planned check of the company’s compute allocation between safety and capability work.
The 1.9x figure is an estimate derived from a particular index and set of models. METR’s confidence interval is broad, and its estimate for Anthropic differs from the smaller acceleration it found across frontier developers overall. The available source material does not establish how much of the change came from specific AI tools, nor how the pace will evolve.
METR said Anthropic’s early-warning threshold may plausibly have been crossed by July 15. Separately, Anthropic wrote in a November 2 SEC filing that it “may imminently meet” its definition of dramatically accelerated progress. Those statements indicate a possible policy threshold; they do not confirm that the threshold was formally triggered or that a specific safeguard was applied.
Policy Response Remains Open
The next developments to watch are whether Anthropic formally determines that its Responsible Scaling Policy threshold has been met and whether it applies additional safeguards as a result. The company’s November 2 filing said it may imminently meet its definition, while METR’s assessment suggests the threshold may plausibly have been reached earlier. The source material does not say what decision Anthropic has made.
Further public detail would help researchers and readers evaluate the estimate, especially evidence currently withheld and the planned analysis of compute allocation. Until more information is available, the report provides an outside estimate based on unusual access to internal data, with important limits on independent verification.
Key Questions
What did METR estimate about Anthropic’s AI progress?
METR estimated that Anthropic’s progress on its measured capabilities accelerated about 1.9 times after October 2025 compared with its earlier pace. Its 95% confidence interval was 1.5x to 2.8x.
What does a 1.9x acceleration mean here?
METR describes a 2x acceleration as making two years of historical progress in one year. Its 1.9x estimate is a comparison of capability progress rates on an index, not a claim that every part of AI development or researchers’ productivity doubled.
How much does METR think AI contributed?
METR estimated that perhaps 80% of the acceleration came from AI use. That is a qualified estimate by the evaluators, not a directly measured causal share.
Has Anthropic confirmed that its policy threshold was crossed?
The supplied material does not confirm a formal threshold determination. METR said a crossing by July 15 was plausible, and Anthropic said in a November 2 SEC filing that it “may imminently meet” its definition of dramatically accelerated progress.
Can the public check METR’s evidence?
Only partly. Anthropic redacted a majority of the evidence METR considered most informative, limiting public review of the basis for the estimate.
Source: Thorsten Meyer AI
