Anthropic launches Claude Fable 5.1 with cheaper cache reads and new privacy controls

Anthropic has launched Claude Fable 5.1 and Claude Mythos 5.1, pairing a performance upgrade with lower costs, revised safeguards and a new answer to enterprise concerns over data retention. Fable 5.1 is broadly available, while Mythos 5.1 remains limited to vetted organizations because it exposes more capability in sensitive areas such as cybersecurity and biology.

The release matters for more than benchmark scores. Anthropic is also changing how some business customers can handle data from its most capable models after pushback over a 30-day retention requirement introduced earlier this year. The company says its planned Enterprise Frontier Safeguards system will keep customer data in cloud infrastructure controlled by the customer rather than Anthropic.

The short answer: Claude Fable 5.1 is Anthropic’s newest generally available frontier model for coding and knowledge work. It keeps the same $10-per-million input and $50-per-million output token prices as Fable 5, but cheaper cache reads cut estimated costs by 25% for typical workloads and by up to about 45% for highly agentic work. Anthropic is also rolling out enterprise privacy controls after customers objected to default 30-day retention for this model class.

What Anthropic announced

Anthropic says Fable 5.1 and Mythos 5.1 are the same underlying model with different safeguard levels. In its official launch announcement, the company describes Fable 5.1 as generally available and Mythos 5.1 as a trusted-access version intended for work where stronger cyber and life-science capabilities are needed.

Detail	Information
Model	Claude Fable 5.1
Availability	Generally available
Mythos 5.1	Restricted trusted access
Input price	$10 per million tokens
Output price	$50 per million tokens
Cache reads	$0.25 per million tokens
Enterprise controls	Enterprise Frontier Safeguards

Fable 5.1 is cheaper without cutting its headline token prices

Anthropic says its pricing changes could reduce typical workload costs by an estimated 25%, with highly agentic workloads seeing savings of up to roughly 45%. 

That distinction matters for developers running long research, coding or agent workflows. These systems often reuse large blocks of context, so cheaper cached input can materially change the cost of a task even when the base token rate stays the same.

Why Anthropic changed its data retention approach

The privacy story is more complicated than a simple return to zero data retention. Anthropic’s default policy for Fable 5.1 requires prompts and outputs to be retained for 30 days for safety monitoring. Anthropic’s privacy documentation says the rule applies to covered models because some misuse patterns can only be detected across multiple requests.

That policy drew concern from enterprise customers that had previously configured zero data retention. CNBC reported on the customer pushback and Anthropic’s decision to develop a different approach with business users. The company says retained data is for safety purposes and is not used for model training.

The replacement is called Enterprise Frontier Safeguards, or EFS. Anthropic says EFS will keep data in cloud infrastructure controlled by the customer while still allowing automated misuse detection. The rollout is due to begin in phases later this fall. Eligible customers can use Fable 5.1 with zero data retention during the transition.

Fable 5.1 and Mythos 5.1 are one model with different access rules

Anthropic is separating capability from access rather than building two unrelated systems. Fable 5.1 and Mythos 5.1 share the same underlying model. Fable adds stronger safeguards, while Mythos gives vetted users access to capabilities that Anthropic considers more sensitive.

The company’s Mythos page says access remains limited to a small set of approved organizations. Anthropic has focused that access on cybersecurity and biology research, where stronger models can have legitimate scientific value but also create higher misuse risk.

Safeguards are less restrictive in some legitimate cyber work

Anthropic reports Fable 5.1 reduces false positives in its cybersecurity safeguards by 60%. The model can now be used to discover software vulnerabilities, although Anthropic still restricts tasks that move toward exploit development. The company’s launch post frames the change as an attempt to stop blocking legitimate security work while maintaining controls on harmful use.

Fable 5.1 enters a competitive AI model pricing and performance market 

The launch also lands in a broader contest between expensive US frontier systems and lower-cost Chinese models. South China Morning Post reported that Fable 5.1 scored 66 on Artificial Analysis’ Intelligence Index, six points ahead of Moonshot AI’s Kimi K3 and Z.ai’s GLM-5.3 in that ranking. The same report noted that Chinese open-weight models continue to compete aggressively on price and distribution.

That creates a more nuanced competition than a simple benchmark race. Anthropic is arguing for stronger frontier performance and enterprise controls, while Chinese developers continue to pressure the market with models that can be materially cheaper to run and, in some cases, openly distributed.

How strong are Anthropic’s benchmark claims?

Anthropic reports large gains over Fable 5 on several internal and public benchmarks. Its launch page lists 52.6% on Terminal-Bench-Science 0.1, 73.4% on CursorBench 3.2.0 and 65.0% on Humanity’s Last Exam with tools. Anthropic also publishes a system card covering capability and safety evaluations for the 5.1 models.

As with any vendor benchmark, those numbers are most useful as directional evidence rather than a guarantee that every real workload will improve by the same amount. Independent rankings and application-specific testing are still important, especially when cost, latency, safeguards and data handling matter as much as raw accuracy.

Who can use Fable 5.1 now?

Fable 5.1 is available to Pro, Max, Team and Enterprise users, and developers can access it through Anthropic’s platform and supported cloud marketplaces. Anthropic also lists availability through Amazon Web Services, Google Cloud and Microsoft Foundry. GitHub says Fable 5.1 is now generally available in Copilot, with zero-retention access available to eligible enterprises during the EFS transition.

What the launch means for enterprise AI buyers

Anthropic’s Fable 5.1 launch reflects the changing priorities of enterprise AI buyers: stronger capability, lower operating costs and greater control over sensitive data. The next challenge will be proving these advantages across real-world deployments.