🔍 Read the full analysis: Can AI Speed Up Its Own Development? What METR Found 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 estimates that Anthropic’s progress on specified AI capability tasks accelerated to about 1.9 times its March 2024–October 2025 rate. The evaluator says AI use likely accounts for much of the increase, but Anthropic redacted most of the evidence METR considered most informative, limiting public scrutiny.
METR, a nonprofit evaluator of frontier AI systems, estimated on November 5 that Anthropic’s progress on specified AI capability tasks is running at about 1.9 times its earlier pace. The assessment, based on 11 weeks of access to internal company data, attributes most of the acceleration likely to AI helping with AI development, but METR said Anthropic redacted a majority of the evidence it found most informative.
METR compared Anthropic’s progress before and after a change point it estimates at October 7, 2025. Its measure, the Anthropic version of the Epoch Capabilities Index, combines results across benchmarks into a single capability score. METR calculated that the score rose by about 15 points per year before the change and about 29 points per year afterward. It describes the result as roughly two years of historical progress occurring in one year.
The report gives a 95% confidence interval of 1.5 to 2.8 times for its 1.9x estimate. METR said it would not be surprised by a result between 1.3 and 3 times under its broader judgment. The estimate is higher than the 1.7x figure in Anthropic’s Haiku 5.5 system card; METR said including newer, unreleased models was the main reason for the difference.
When METR examined frontier models across developers rather than Anthropic alone, its estimate was smaller: progress rose from about 17 to 24 index points per year, or 1.4 times. The report says this could reflect Anthropic catching up from a weaker position in 2025, which would make its company-specific acceleration look larger, or differences in how earlier models performed on benchmarks. Varying assumptions about task difficulty shifted the Anthropic estimate from 1.1 to 2.9 times, while most other checks remained near 2 times.
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
AI’s Role in Anthropic’s Faster Progress
The findings bear on whether AI tools are beginning to speed up the work that produces more capable AI systems. METR’s estimate is not a finding that AI systems are independently building their successors: the report concerns AI assisting human research and engineering, with people still directing the work. Anthropic defines recursive self-improvement as an AI system autonomously designing and developing its successor and says that stage has not been reached.
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 attribution is an estimate, not a directly measured share. If AI assistance is materially increasing the rate of research progress, it could affect how quickly model capabilities develop and how much time companies and policymakers have to respond. The report does not establish how that pace will evolve or what it means for future systems.
Anthropic’s Responsible Scaling Policy sets a 2x progress-rate threshold as one possible trigger related to automated research and development, when acceleration is plausibly substantially attributable to automation. METR said it is plausible the threshold was crossed by July 15, 2026. Anthropic wrote in a November 2 SEC filing that it “may imminently meet” its definition of dramatically accelerated progress. The two statements describe a potential policy threshold, not proof that a specific safety condition has been met.
Products Worth Considering
Apple MacBook Pro MPHF3LL/A 14.2-Inch Laptop | M2 Pro 12-Core Chip 16GB Unified Memory 1TB SSD | 19-Core GPU Liquid Retina XDR Display Space
Apple MacBook Pro 14.2-inch with M5 Max, 18-Core CPU, 20-Core GPU, 48GB Memory, 2TB SSD, Space Black
Apple MacBook Pro 16.2-inch with M5 Pro, 18-Core CPU, 20-Core GPU, 48GB Memory, 1TB SSD, Nano-Texture Display, Space Black
As an affiliate, we earn on qualifying purchases.
How METR Measured the Change
METR’s assessment differs from an outside test of a publicly released model. Its staff worked with Anthropic’s internal information, including benchmark scores for unreleased models, training metrics, surveys and about 15 interviews with company staff. METR said about 2.5 full-time-equivalent staff worked on the assessment from August 12 to October 29, using Anthropic devices. The evaluator said it took no payment; Anthropic provided laptops and free tokens and retained the right to redact non-public information.
The measure tracks progress on benchmarked capabilities, not every part of AI research and development or the overall productivity of Anthropic’s workforce. METR explains acceleration as the rate at which capability progress compares with an earlier baseline: a 2x rate means that progress historically achieved in two years occurs in one year. That is distinct from saying researchers feel twice as productive, because other stages of development still take time.
Anthropic’s own figures offer another view of AI’s role in its work. The source report says the company measured Claude as leading 26% of its AI research and development work in August 2026, compared with less than 1% in February, with about 30,000 agents working at a time on its main internal platform. These are company measurements reported in the source material; the available details do not establish how “leads” was defined or independently validate the figures.
““roughly two years of historical progress over the last year””
— METR
What the Redactions Leave Unknown
The public cannot fully inspect METR’s analysis because Anthropic redacted a majority of the evidence METR found most informative. The source also says a planned examination of how the company divides compute between safety work and capability development was dropped. METR’s conclusion is therefore based on access that the public does not share, and readers cannot independently assess all of the underlying records or reproduce the analysis from the material described.
The precise contribution of AI tools remains uncertain. The estimate that they account for perhaps 80% of the acceleration is METR’s judgment; the report’s supplied summary does not present it as a direct causal measurement. The estimate also depends on benchmark selection and assumptions about task difficulty. The smaller 1.4x result across developers underscores that the Anthropic-specific figure should not be treated as a measure of the entire AI industry.
It is also not clear from the source material whether Anthropic has formally determined that its policy threshold has been crossed, what safeguards would follow if it did, or whether the reported rate will persist. METR’s plausible July 15 threshold date and Anthropic’s qualified SEC filing do not settle those questions.
Policy Thresholds and Further Evidence
The next public developments to watch are whether Anthropic says it has met the Responsible Scaling Policy’s automated R&D criteria and whether it reports any resulting safeguards. The policy identifies both a model capable of replacing the company’s research staff at competitive cost and a doubling of progress plausibly substantially attributable to automated research or engineering as possible triggers.
Further assessments could clarify whether the estimated acceleration holds across more models, benchmarks and AI developers, and whether the rate changes over time. For now, METR’s report provides a detailed external estimate based on internal access, but the redactions and methodological uncertainties limit what outsiders can verify. No additional publication date or formal policy decision is specified in the source material.
Key Questions
What did METR find about Anthropic’s AI development pace?
METR estimated that Anthropic’s progress on its benchmark-based capability measure accelerated to about 1.9 times its earlier rate. It calculated a rise from roughly 15 to 29 index points per year after an estimated October 7, 2025, change point.
Does the report say AI is developing its own successor?
No. The report describes AI helping people with AI research and engineering. Anthropic’s definition of recursive self-improvement involves an AI system autonomously designing and developing its successor, a stage Anthropic says has not been reached.
How much of the acceleration did METR attribute to AI?
METR estimated that perhaps about 80% may be due to AI use. That is the evaluator’s judgment, not a directly measured causal share, and the underlying evidence is partly unavailable to the public.
Why is the 1.9x estimate uncertain?
METR’s estimate changes with assumptions about task difficulty, ranging from 1.1 to 2.9 times in the checks described. Its cross-developer estimate was lower, at 1.4 times, and Anthropic redacted much of the evidence METR found most informative.
Has Anthropic confirmed it crossed its 2x policy threshold?
The supplied material does not confirm that Anthropic formally determined the threshold was crossed. METR said crossing it by July 15 was plausible, while Anthropic’s November 2 filing said it “may imminently meet” its definition of dramatically accelerated progress.
Source: ThorstenMeyerAI.com
