Fable 5.1 is designed to handle end to end programming jobs - entire software projects and full codebase reviews - with stronger follow-through. On the science front, it is built to help design experiments, simulate scenarios, and read dense diagrams and tables more effectively.
Anthropic plans to release Fable 5.1 on Tuesday. It also expects to make an updated Mythos 5.1 available the same day to a limited set of invite-only users. While the two models are closely related, Fable is blocked from carrying out some cybersecurity and science-related tasks.
Enterprise pricing per token stays the same as Fable 5, but Anthropic is slashing what it charges when the model resurfaces information it has already processed. That step gets a 75% price cut, which the company says should produce a "substantial reduction" in total run costs.
For lengthy, intricate assignments, the resurfacing step often represents half - or even the majority - of total tokens, according to Dianne Penn, the head of product management for research and labs. "This was one area we wanted to make improvements on, responding to customer feedback," she said. Penn added that the model aims to complete tasks while consuming fewer tokens overall.
Even when headlines change quickly, patient saving and steady contributions pay off, grab the free Always Be Buying E-Book
Safety tooling is getting sharper too. Fable 5.1's classifiers that detect and block problematic prompts are described as more accurate with far fewer false positives, which should help practitioners in cybersecurity and biology-focused workflows move more quickly.
Anthropic said it plans to let business customers keep data on their own cloud infrastructure when using its most capable models, while still running Anthropic's safety checks on that data. The company called it a stark shift from its earlier retention policy intended to limit cyber risk, confirming prior Bloomberg reporting.
US government safety concerns about both Fable and Mythos led to access restrictions in June. After Anthropic added classifiers that better spot and block cybersecurity tasks, limits on Fable 5 were lifted. Later, Mythos 5 access returned, though only a narrower subset of users received it.
Companies are trying to prove the value of AI while keeping bills in check, even as lower cost, open-weight models from China keep improving and turn up the heat on US players. Over the past year, Anthropic and OpenAI have been building tools to streamline tasks across software development, security, finance, the sciences, and healthcare to win more business customers and justify big valuations. Bloomberg reports that Anthropic expects annualized revenue to exceed $65 billion, a figure over seven times what it was tracking at by the end of 2025. Anthropic and OpenAI alike have submitted confidential filings for public listings, with Anthropic potentially debuting as early as this fall. For anyone watching how fast AI shows up in everyday products, cheaper runs and more flexible data control are the sort of tweaks that can speed up adoption where you actually work.