🔍 Read the full analysis: Is AI’s Own Development Picking Up Pace? METR Examines Anthropic on ThorstenMeyerAI.com
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
TL;DR
METR’s November 5 assessment estimates that Anthropic’s progress on measured AI capabilities accelerated to about 1.9 times its March 2024–October 2025 pace. The nonprofit believes AI use may account for much of the increase, but says Anthropic redacted most of the evidence it found most informative, limiting public scrutiny.
METR, a nonprofit that evaluates frontier AI systems, estimated on November 5 that Anthropic’s measured model progress is advancing about 1.9 times its earlier pace, based on an assessment using internal company data. METR said AI tools likely account for much of the acceleration, but also reported that Anthropic redacted a majority of the evidence it considered most informative, leaving important parts of the analysis difficult for outsiders to verify.
METR compared Anthropic’s progress on its capability index before and after an estimated change point of October 7, 2025. The nonprofit calculated an earlier trend of about 15 index points per year and a later trend of roughly 29 points per year, producing its central estimate of 1.9 times the previous rate. METR gave a 95% confidence interval of 1.5 to 2.8 times and said it would not be surprised by estimates from 1.3 to 3 times.
The assessment used benchmark results for released and unreleased models, training metrics, surveys and about 15 interviews with Anthropic staff. METR said about 2.5 full-time-equivalent staff worked on the assessment over 11 weeks, using Anthropic-provided devices. It said it received no payment; Anthropic supplied laptops and free tokens and retained the right to redact non-public material.
METR’s interpretation is that AI assistance, rather than faster growth in training compute, explains most of the change. The nonprofit estimated that perhaps about 80% of the acceleration came from AI use, while describing that figure as an estimate. Separately, Anthropic’s own measurement says 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 operating at a time on its main internal platform; those figures are company-reported, not METR’s independent measurement.
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 finding bears on whether AI systems are beginning to speed up the research and engineering used to create more capable systems. If AI assistance makes model development faster, the pace of capability gains and the time available to evaluate risks could change. But METR’s estimate concerns progress on well-specified benchmark tasks; it does not establish that AI systems can independently design and build their successors.
The distinction matters for interpreting “recursive self-improvement.” In the description provided by Anthropic, that would mean an AI system autonomously designing and developing its successor. The current picture in the report is instead one of AI tools helping human researchers, who still set goals and make decisions. METR’s results may indicate that this assistance is already affecting development speed, but they do not show that the process is fully autonomous or that future acceleration will continue at the same rate.
The limits on disclosure also matter. A third-party assessment can provide a useful check on a company’s own accounts, but redactions make it harder for other researchers to inspect assumptions, evidence and possible alternative explanations. The report therefore offers a significant estimate without providing the public with the full basis needed to independently reproduce it.
Popular right now
De'Longhi La Specialista Arte Evo Espresso Machine with Grinder
Breville Bambino Espresso Machine, Brushed Stainless Steel
Dyson V15 Detect Cordless Vacuum
As an affiliate, we earn on qualifying purchases.
Anthropic’s Policy Threshold
Anthropic’s Responsible Scaling Policy, version 3.4, took effect on July 8, 2026. It identifies two possible triggers for safeguards related to automated research and development: a model capable of replacing the company’s research staff at competitive cost, or a doubling in the pace of progress plausibly attributable in substantial part to automated research or engineering. The policy defines a doubling as making in one year the progress that would otherwise take two years.
Anthropic describes that threshold as an early warning rather than proof that severe risks are present. In a November 2 filing with the U.S. Securities and Exchange Commission, the company wrote that it “may imminently meet” its definition of dramatically accelerated progress. METR said it considered it plausible that the policy’s early-warning threshold had been crossed by July 15, although that timing is an assessment, not a confirmed determination by Anthropic.
METR’s estimate is not the same as a claim that every part of AI development has doubled in speed. It measures a particular pattern in capability-index results. When METR examined frontier models from all developers rather than Anthropic alone, the estimated pace change was smaller: from about 17 to 24 points per year, or 1.4 times. That difference leaves open how much the Anthropic-only estimate reflects its own acceleration and how much reflects its position relative to the broader frontier.
““A 2x acceleration means that 2 years of historical progress is made in 1 year.””
— METR
Evidence Behind the Estimate
METR said Anthropic redacted a majority of the evidence it found most informative. The report’s public account therefore does not allow readers to inspect all of the underlying data. METR also dropped a planned check of how Anthropic divides compute between safety work and capability development, so that question remains unanswered in this assessment.
The 1.9-times result depends on choices about the data and measurement. METR said alternative assumptions about task difficulty moved its estimate from 1.1 to 2.9 times, although it reported that most other checks stayed near 2 times. The estimate was also higher than the 1.7-times figure in Anthropic’s Haiku 5.5 system card; METR attributed much of the difference to its inclusion of newer, unreleased models.
METR offered two possible explanations for the gap between Anthropic’s estimate and the broader frontier estimate: Anthropic may have been behind the frontier in 2025, which could make its later rise look larger, or its earlier models may have been less tuned to benchmarks, which could make the increase look smaller. The available material does not settle which explanation is more accurate. It also does not establish that the reported pace will persist or translate into equivalent gains across all AI research tasks.
Further Disclosure and Policy Checks
The next useful developments would be additional public detail from Anthropic or METR about the redacted evidence, measurement choices and the omitted compute-allocation analysis. Without those details, outside researchers have limited ability to test the assessment’s main estimate or determine how much the result depends on unpublished information.
Anthropic’s own policy status is also unresolved in the material available here. Its SEC filing said the company may imminently meet its definition of dramatically accelerated progress, while METR judged that an early-warning threshold may plausibly have been crossed by July 15. Neither statement, by itself, confirms that Anthropic formally determined the threshold was met or specifies what safeguards, if any, it has activated.
For now, the report provides an outside estimate of faster measured progress and a qualified explanation that AI assistance may be a major contributor. Whether that acceleration is robust across other measures, how much comes from AI tools, and what it means for the timing of safeguards remain open questions.
Key Questions
What did METR conclude about Anthropic’s development pace?
METR estimated that Anthropic’s measured capability progress is advancing at about 1.9 times its rate before an estimated October 2025 change point. Its 95% confidence interval was 1.5 to 2.8 times.
Does the report show that AI is building its own successor?
No. METR’s assessment concerns AI tools assisting human research and development. It does not establish that a system can autonomously design and develop its successor.
Why does METR think AI accounts for much of the acceleration?
METR estimated that perhaps about 80% of the increase was due to AI use and said it was confident the change did not come from faster growth in training compute. The 80% figure is an estimate, not a directly measured attribution.
How much of METR’s evidence is public?
METR said Anthropic redacted a majority of the evidence the nonprofit found most informative. The redactions limit what outsiders can independently check.
Has Anthropic confirmed that it crossed its policy threshold?
Not in the source material. Anthropic said in a November 2 filing that it may imminently meet its definition of dramatically accelerated progress. METR said crossing an early-warning threshold by July 15 was plausible, but that is not confirmation of a formal company determination.
Source: ThorstenMeyerAI.com
