Overall price reduction of 75%! Anthropic released Fable 5.1, and its annualized revenue soared seven times, impacting its fall IPO

The Zhitong Finance App has learned that Anthropic PBC is releasing a new version of its powerful Fable artificial intelligence model. The company said the new version performs better in coding and scientific tasks, while also being more affordable — all of these changes are in response to customer feedback.

The company said Fable 5.1, which is scheduled to be released on Tuesday, is better at performing long and complex computer programming tasks, such as software projects or code reviews that require work across an entire application codebase. Anthropic said the new version can also better assist scientific work, such as proposing experimental designs, simulating what might happen during experiments, and reading information in dense charts and tables.

At a time when open source, low-cost models are rapidly advancing in China, companies are increasingly trying to demonstrate and control the cost of using AI. These trends are putting pressure on American model makers such as Anthropic.

For Fable 5.1, the price per token charged by Anthropic to enterprise users will be the same as Fable 5, but the fee charged to users when the AI model reloads blocks that have been previously processed will be reduced.

Anthropic said the 75% price reduction offered during this process would contribute to a “significant reduction” in the overall cost of running the model.

Diana Payne, who is responsible for the company's research and laboratory product management, said that in some of the longer and more complex tasks people run on Anthropic's AI model, this process may take up half or more of the total amount of tokens used.

“This is an area where we want to improve in response to customer feedback.” she said.

Payne added that the model also aims to perform tasks better by using fewer tokens.

The company said Fable 5.1 also improved the safety classifier — the software used to identify and block illegal instructions issued to models. They are now more accurate and the false alarm rate has been drastically reduced, which means users working on cybersecurity and biological-related tasks will be able to work more efficiently.

Anthropic added that it plans to allow commercial customers to keep data on the customer's own cloud computing infrastructure rather than Anthropic's cloud while still running the company's security checks on this data as they use its most powerful AI models. This marks a major change to previous data retention policies aimed at mitigating potential cyber risks, and confirms what Bloomberg reported earlier.

Anthropic is also expected to release an updated version of its invitation-only Mythos 5.1 model to a small pre-selected group of users on Tuesday. Although Fable is similar to Mythos, Fable is prohibited from performing certain cybersecurity and science-related tasks.

The US government's security concerns about the capabilities of these two AI models led it to restrict access to the two models in June; restrictions on the models were lifted after Anthropic added a classifier that allowed Fable 5 to identify and block more cybersecurity tasks. Access to Mythos 5 was later restored, but limited to a smaller user base.

For most of the past year, Anthropic and rival AI model maker OpenAI have been developing AI tools to simplify a wider range of professional tasks (including programming, cybersecurity, financial services, science, and healthcare), with the goal of reaching more enterprise customers and supporting their high valuations. As reported by the media, Anthropic's annualized revenue is expected to surpass $65 billion; compared to the company's pace at the end of 2025, this figure is more than seven times higher.

Both Anthropic and OpenAI have submitted confidential documents to begin the listing process; Anthropic is expected to go public as soon as this fall.