Claude Fable 5.1 and Mythos 5.1: What Changes at Anthropic

Anthropic opened September 2026 with two releases at once: Claude Fable 5.1, its new general-purpose model, and Claude Mythos 5.1, a more capable version available only through the company’s trusted-access programs. Anthropic calls the pair “the world’s most advanced models for coding and knowledge work” — and, for the first time, the headline is not just performance. It is cache pricing and access control.

Quick answer: what changed?

Fable 5.1 keeps Fable 5’s list price ($10 per million input tokens, $50 per million output tokens) but cuts cached input reads from $1.00 to $0.25 per million. In practice, typical workloads become roughly 25% cheaper and heavily agentic workloads save up to 45%. Mythos 5.1 is restricted to trusted-access programs focused on cybersecurity and life sciences.

Claude Fable 5.1: cheaper where it matters

The strongest argument for Fable 5.1 is not a benchmark; it is the monthly bill. Teams running Claude inside agents — workflows where the model re-reads the same context dozens of times — pay mostly for cache reads. Cutting that line item by 75% changes the economics of many products.

The technical limits remain generous: 1 million tokens of input context and 128K tokens of output. Anthropic also shipped three breaking API changes, including the removal of forced tool choice. If you have an integration in production, review your code before switching models.

Claude Mythos 5.1: the model that is not for everyone

Mythos 5.1 is the most intriguing part of the announcement. It does not appear in the public API. Anthropic offers it only through trusted-access programs for cybersecurity and life-sciences work, inside a package called Enterprise Frontier Safeguards — which combines zero data retention with misuse-detection systems.

Task boundaries per model

Anthropic defined clear usage rules: Fable 5.1 may identify vulnerabilities, but penetration testing, exploit generation and binary-based vulnerability scanning are redirected to the Opus models within the controlled programs. It is a sign that labs are now segmenting capability by risk, not only by price.

Price comparison (API, per million tokens)

Model	Input	Output	Cache read	Context
Claude Fable 5	$10	$50	$1.00	1M
Claude Fable 5.1	$10	$50	$0.25	1M
GPT-6 Astra (OpenAI)	$10	$50	$1.00	1.05M
Gemini 3.8 Flash (Google)	$0.75*	$3.75*	—	1M

*Promotional pricing through December 31, 2026; then $1.50 / $7.50.

Why this matters to you

If you are a developer or run an AI-powered product, the cache price is the number to put in your spreadsheet: agents that reuse context now cost close to half. If you use claude.ai, the change shows up as more consistent answers on long tasks. And if you work in security, the message is that the most capable models will require credentials and access programs — a credit card is no longer enough.

Frequently asked questions

Did Claude Fable 5.1 get more expensive?

No. The list price is the same as Fable 5 ($10 / $50 per million tokens). What changed is cached input reads, down from $1.00 to $0.25.

How can I use Claude Mythos 5.1?

Only through Anthropic’s trusted-access programs for cybersecurity and life sciences. There is no public API access.

Do I need to change my code to migrate?

Possibly. Anthropic announced three breaking API changes, including the removal of forced tool choice. Test in a separate environment before swapping the model in production.

At DigitalRadar, we track every release in the AI race. Stay on the radar so you do not miss the next update.