AI is now helping to build better AI, and that change may be happening faster than many people expected. A new assessment from METR puts a number on that shift — and explains why the number needs careful reading.
Key takeaways
- METR estimates a 1.9-times increase in the measured pace of progress on Anthropic's internal benchmark index since October 2025.
- The 95% confidence interval is 1.5 to 2.8 times, so the estimate comes with real uncertainty.
- Accelerated growth in training compute did not explain the measured improvement.
- Much of the underlying evidence was redacted, so readers cannot check every detail.
- The roughly 100-page assessment was released on 5 November 2026.
What did METR look at?
METR released a high-access assessment on 5 November 2026, asking a hard question: is AI accelerating research into better AI at Anthropic?
The investigators worked from 12 August to 29 October 2026, using internal measurements, researcher interviews, and technical context from inside Anthropic.
What is a benchmark index?
A benchmark is a set of tests used to measure performance on a defined task. A benchmark index brings several tests together into one combined measure.
Think of it like a school report card: one test covers one subject, while a report card combines several subjects into one broader picture. An index shows movement on the tests it includes, not every kind of progress.
What does 1.9x faster actually mean?
METR estimates a 1.9-times increase in the measured pace of progress since October 2025. In plain words, work that once moved the index forward at one speed is now, by this measure, moving about 1.9 times as fast.
It does not mean every AI task is now 1.9 times faster, or that models are 1.9 times smarter in every way. It means the pace on this one internal index has increased by that estimated amount. The 95% confidence interval of 1.5 to 2.8 times shows the estimate is not exact.
Why more computing power is not the whole story
One obvious explanation would be more computing power. Yet METR found that accelerated growth in training compute did not explain the measured improvement. That points to other factors, but it does not, on its own, identify exactly which ones mattered most.
Why so much was redacted
Redaction means parts of a document are hidden before publication. Here, much of the underlying evidence was redacted from the roughly 100-page assessment.
Internal research can include private methods and data a company does not want to share. Still, redaction matters for trust: when evidence is hidden, readers cannot repeat the checks themselves. That does not mean the estimate is wrong — it means readers must trust METR's judgement and access, rather than verify each step directly.
What this does not prove
This assessment does not prove that AI has achieved superintelligence, that AI research will keep speeding up at the same rate, or that the same change is happening at every AI company.
It is a focused study, based on one company's internal benchmark index, with the comparison starting in October 2025. It also does not prove that faster internal research will automatically lead to better tools for everyday users. A faster research pace is news about how AI is being built, not a promise that your next AI tool will be twice as useful.
For practical uses today, see How AI Helps You Work Faster Every Day, The AI Spending Boom Faces Its Reality Check, and How Small Shops Can Use AI Without Spending Money.
Conclusion
METR's assessment suggests AI research at Anthropic is moving faster, by an estimated 1.9 times on an internal benchmark index since October 2025, with a 95% confidence interval of 1.5 to 2.8 times. Accelerated growth in training compute did not explain the improvement, and much of the underlying evidence was redacted.
That is a careful, limited claim — not about superintelligence, and not a promise that progress will continue at the same rate. Stay curious and stay careful: follow the evidence, note what is still hidden, and judge new AI tools by what they do in real use.
FAQ
What is METR?
METR is an independent evaluator. It released a high-access assessment on 5 November 2026 asking whether AI is accelerating research into better AI at Anthropic.
What does a 1.9x increase in the pace of progress mean?
It means METR estimates progress on Anthropic's internal benchmark index is moving about 1.9 times as fast as in October 2025. The 95% confidence interval is 1.5 to 2.8 times, so the exact size is uncertain.
Why does redaction matter in this assessment?
Much of the underlying evidence was redacted, so readers cannot check every detail. That does not prove the estimate is wrong, but readers are relying on METR's judgement and access, rather than on a fully open record.
Sources
- Chris Abraham, "AI Roundup: The Work Is Speeding Up" — https://chrisabraham.net/2026/11/ai-roundup-the-work-is-speeding-up.html
- The Neuron, "Everything That Happened in AI Today — Friday, November 6, 2026" — https://www.theneuron.ai/digest/everything-that-happened-in-ai-today-friday-november-6-2026/
- SXF, "Brief" — https://sxf.si/brief/
Researched with AI assistance and reviewed for accuracy.
Follow on WhatsApp
For more simple AI tips, follow our WhatsApp channel: https://www.whatsapp.com/channel/0029Vb9XzdHDJ6H4TFNAl61r

Comments
Post a Comment