Fable 5.1 Cuts Cache Costs as Much as 75%

Anthropic released Fable 5.1 on Tuesday, and the news is not the model. It is the two things the company gave up to ship it.

The model is better at long programming jobs — software projects and code reviews that span an entire application — and at scientific work like designing experiments, simulating outcomes, and reading dense diagrams and tables, Bloomberg reported. Enterprise customers pay the same per-token price as before.

They will pay far less overall. Anthropic cut the price of cached tokens — the material a model resurfaces from work it already processed — by 75 percent. Dianne Penn, who leads product management for research and labs, told Bloomberg that cached tokens can account for half or more of the tokens consumed on long tasks. Bloomberg put the reason plainly: open-weight models from China are improving fast and cost less, and that is pressure on American labs.

This desk has been tracking that pressure for two weeks — Zhipu’s weights going free, the stealth model that burned through trillions of tokens before revealing itself as Chinese. Today it showed up on an American price sheet.

The second concession is bigger. Anthropic will let business customers keep their data on their own cloud infrastructure rather than Anthropic’s, while still running the company’s safety checks against it. Bloomberg calls it a stark reversal of a retention policy built to contain cyber risk. In June, the U.S. government restricted access to both Fable and its invite-only sibling Mythos over safety concerns; the limits came off Fable 5 after new classifiers were added, and Mythos 5 returned to a smaller group. Fable 5.1’s classifiers throw fewer false positives, which the company says will let security and biology researchers work without tripping the guardrails.

Anthropic is tracking to more than $65 billion in annualized revenue, per Bloomberg, over seven times its pace at the end of 2025. It has filed confidentially to go public, possibly this fall.

A week with it, already

The team at Every had the model under embargo and has published its review:

“We Tested Anthropic’s Fable 5.1 for a Week.” Video: Every.