METR says Anthropic is improving its models about twice as fast as before, mostly thanks to AI. Plan your automations for faster model turnover

The models under your automations are about to change more often, if METR's read on Anthropic is right.

What happened

From August 12 to October 29, METR ran a high access evaluation inside Anthropic to measure how much AI is speeding up AI research. Its November 5 report concludes that, as of October, Anthropic is likely improving its models on well specified tasks at around twice the rate it did between March 2024 and October 2025. METR stresses high uncertainty: the true figure could be closer to 1.3x or 3x.

The Verge reports METR's view that the majority of the acceleration is due to AI, mostly AI assisting human researchers rather than working alone, though METR sees signs that autonomous contributions will grow. Employees using Anthropic's most capable internal model self reported being around 12 times as productive. METR also says Anthropic has finished training three models in the past six months that it has not released.

There are real caveats. Anthropic redacted a majority of the evidence METR found most informative, and METR says the public has to rely on its judgment more than would be ideal. Anthropic cofounder Sam McCandlish said the redactions protect commercially sensitive details, and that the company will publish its own report on the speed of AI progress in the coming weeks.

My take

I won't weigh in on the safety debate here. The practical point for anyone running AI inside business workflows is simpler: if one major lab is moving twice as fast, model releases and retirements will come faster too.

That changes how I build:

1. Keep the model name in one config value, never scattered across workflows, so a swap is a one line change.
2. Keep a small test set for every AI step, say twenty real inputs with known good outputs, and run it before switching models.
3. Pin versions for production and test new ones on a copy first.
4. Log which model handled each record, so when output changes you can trace why.

Faster progress is good news for clients only if the system can absorb it without breaking.