METR says Anthropic is likely improving its models faster

The models are moving faster

METR says Anthropic is likely improving its models on well-specified tasks at around twice the rate it did between March 2024 and October 2025, and it gives that estimate with high uncertainty. METR also says a majority of that acceleration is likely attributable to Anthropic's extensive use of AI to do R&D.

We can't see all the evidence

Anthropic redacted a majority of the evidence METR found most informative, and METR says the public has to rely on its judgment much more than would be ideal. For anyone building on these models, that is a reason to run their own checks.

Claude steps into personal finance

Anthropic announced Personal Finance in Claude, where users connect their bank and credit card accounts through Plaid and ask Claude questions about their finances. Anthropic says access is read-only, meaning it can't move, manage or invest users' money, and it would like to hear from banks, wealth managers and financial services companies who want to help shape it.

Shopping agents still fail often

Visa put 22 AI models from 12 labs through realistic shopping scenarios, and 57.2% of the test runs passed while 42.8% failed. Among the best-performing models 82.2% of runs passed, against 28.9% among the worst.

Agents with people in the loop

Lacerte, BILL's CEO, said Robin sends each case to the right underwriter, pulls the customer's data into a summary and provides a one-click link to make a decision. He said each case and customer wait time now takes about 12 minutes instead of three hours. If you work in credit or operations, which of your queues looks like that, and what would a benchmark like Visa's say about it?