Anthropic launches Fable 5.1: Up to 75% price reduction, enhanced programming and research capabilities

Anthropic released a new version of its flagship AI model, Fable 5.1, on Tuesday, significantly reducing costs for certain use cases while strengthening its programming and research capabilities. This move is a direct response to price pressure from enterprise clients and also reflects intensifying market competition among US AI model providers.

The core change in the new version centers on pricing strategy adjustments: Anthropic has reduced the fee for model cache reads—reprocessing previously handled information—by 75%. The company stated this will result in a “substantial reduction” to overall operational costs.

Dianne Penn, Head of Product Management at the company, pointed out that this aspect often accounts for half or more of token consumption in complex, long-running tasks, making it one of the most commonly cited pain points among customers.

Meanwhile, Anthropic also announced it will allow enterprise clients to retain their data on their own cloud infrastructure while still being subject to Anthropic’s security reviews. This policy marks a major shift from the company’s previous data retention requirements.

Additionally, the invitation-only advanced model, Mythos 5.1, will be rolled out on a limited basis the same day to a select group of pre-chosen users.

Fable 5.1 keeps the same pricing for new input tokens as its predecessor, Fable 5. The primary rate change focuses on “cache reads”—charges incurred when the model rereads previously processed content. This fee will drop by 75%.

Dianne Penn stated that this aspect frequently accounts for more than half of the token usage in tasks like software development and code review, which require model traversal of an entire application’s codebase. “This is an area we wanted to improve and a response to customer feedback,” she said.

Beyond reduced pricing, the new model itself has been optimized to accomplish the same tasks with fewer tokens.

The context for these changes is clear: Model competition is becoming ever fiercer; at the same time, enterprise buyers are growing increasingly cautious about AI expenditures and urgently need a more predictable cost structure.

In terms of performance, Fable 5.1 focuses on enhancing two types of professional scenarios.

For programming, the new model is better equipped to handle long-chain tasks that span entire codebases, covering large-scale software project development and comprehensive code reviews. In research, the model also performs better in designing experimental protocols, simulating experiment processes, and reading high-density chart data, among other tasks.

The security classifier has also been simultaneously optimized—with a significant decrease in false positives. This means users in cybersecurity and biology-related fields will experience fewer unnecessary blocks by the system, which should improve operational efficiency.

Notably, the US government had previously imposed access restrictions on the Fable and Mythos models in June over security concerns. After Fable 5 incorporated classifiers capable of recognizing and blocking more cybersecurity tasks, the restriction was lifted; access to Mythos 5 was also resumed, but remains available to a much smaller user group.

The company’s newly-announced data retention policy adjustment may be one of the most attractive changes for enterprise clients in this update.

The new policy allows enterprise users working with Anthropic’s most powerful models to keep their data within their own cloud computing infrastructure, rather than being required to store it on Anthropic’s servers—while Anthropic will still conduct security checks on the data in question.

The previous policy was designed to reduce cybersecurity risks, but it also deterred some enterprise clients who have strict data sovereignty requirements.

The release of Fable 5.1 is part of Anthropic’s broader strategy—competing with OpenAI to expand its enterprise customer base and cover professional scenarios like programming, cybersecurity, financial services, science, and healthcare.

Anthropic expects annualized revenue to surpass $65 billion, more than seven times its operating pace at the end of 2025.

Both Anthropic and OpenAI have submitted confidential IPO filings, with Anthropic potentially going public as early as this fall.

As the IPO window approaches, continuously expanding market share among enterprise clients and converting high valuations into visible revenue growth will be crucial for both companies.