Claude Fable 5: price, access and Fable 5.1

Claude Fable 5: price, access and Fable 5.1

Claude Fable 5 is Anthropic's most powerful publicly available AI model, and it is available in Claude.ai, Claude Code, Claude Cowork and through the API. It launched on 9 June 2026, was shut off worldwide after three days and came back on 1 July after 19 days offline. Since 1 September 2026 there is also a successor, Fable 5.1. Here is what the model is, what it costs, how to get access, what happened in June and how Mythos 5 fits in.

Looking for the Xbox game Fable? This is about Anthropic's AI model, not the game series with the same name.

What is Claude Fable 5?

Fable 5 is Anthropic's new flagship model, built for demanding reasoning and long, autonomous tasks. It is the first model of the fifth generation and belongs to the so-called Mythos class. It handles codebases, analytical work and tasks that run for days without losing the thread.

The launch numbers were staggering. Stripe reported that Fable 5 compressed months of work into days and migrated a 50-million-line codebase in a single day, something a whole team would otherwise have needed two months for. Fable 5 also took top scores on several demanding benchmarks, including Cognition's FrontierCode and Hebbia's finance benchmark.

To compare it with Anthropic's other models (Opus 5.5, Sonnet 5.5 and Haiku 5.5), we have collected pricing, context windows and benchmarks in our guide to Claude.

Fable 5.1 — what is new?

Fable 5.1 is an update to Fable 5 that Anthropic released on 1 September 2026. According to Anthropic it is better at coding, knowledge work and problem-solving, and the company highlights research and scientific work in particular. Fable 5.1 is available in the same places as Fable 5: Claude.ai, Claude Code, Claude Cowork, the API and through AWS, Google Cloud and Microsoft.

Three things differ in practice:

• Per-token pricing is unchanged, but cache reads became 75% cheaper. Anthropic says the cost of typical workloads ends up about 25% lower, and up to 45% lower for agentic tasks.
• Fewer unnecessary blocks. Anthropic reports 60% fewer false positives in cybersecurity and 85% fewer interventions on benign biology questions.
• Data retention. Anthropic is introducing Enterprise Frontier Safeguards in autumn 2026. Until then eligible customers can get zero data retention.

How you reach 5.1 on your plan is covered below. According to Anthropic's help centre, 5.1 works the same way as Fable 5 on each plan.

How much do Fable 5 and Fable 5.1 cost?

Through the API, both Fable 5 and Fable 5.1 cost $10 per million input tokens and $50 per million output tokens. Prompt caching gives up to 90% off input, and with 5.1 cache reads dropped to $0.25 per million tokens. If you choose US-only inference, a 1.1x multiplier applies.

In the subscriptions, your plan decides:

• Max and Team Premium: included in the plan, up to 50% of the weekly usage limit.
• Pro and Team Standard: through usage credits (pay as you go). Existing users received a one-time $100 credit when the change was made.
• Free: the model is not available.

Fable 5 is the most expensive in the model family. For many workflows Opus 5 or Sonnet 5 is plenty, and they cost a fraction.

How to get access to Fable 5 today

Since 20 July 2026 access has been permanently split as follows, according to Anthropic's help centre:

• Max and Team Premium: pick Fable 5 in the model selector.
• Pro and Team Standard: enable usage credits first, otherwise the model does not show up as an option.
• API (Claude Platform): choose the model directly in the call.
• Cloud platforms: AWS, Google Cloud and Microsoft Foundry.

The model is usable in Claude.ai, Claude Code, Claude Cowork and through the API. At the July redeployment, questions about biology and cybersecurity were automatically routed to the Opus models for safety reasons, and you were not charged Fable 5 prices for those calls. Fable 5.1 has fewer such blocks.

What is Claude Mythos?

Claude Mythos is the same kind of model as Fable with fewer safeguards, and it is not open to the public. Anthropic first presented Claude Mythos Preview in April 2026 within Project Glasswing, a programme for finding vulnerabilities in critical software. The model was considered so capable that it was not released publicly, and the headlines were about it finding decades-old security holes.

Mythos 5 arrived alongside Fable 5 on 9 June: the same underlying model, but with some safeguards removed for cybersecurity experts and critical infrastructure. Mythos 5.1 followed with Fable 5.1. You can only get it through Anthropic's verification programmes, the Cyber Verification Program for defensive work and the Life Sciences Verification Program for research, and for now only for organisations in the US.

The difference in short: Fable is the model you can use, Mythos is the same model for verified experts.

Why was Fable 5 shut down?

This is the short version of the 19 days offline.

12 June — the directive

Late on Friday 12 June, at 17:21 ET, Anthropic received a directive from the U.S. government citing national security: block access for all foreign nationals, both inside and outside the US. Because the company cannot separate foreign users from the rest in real time, the result was a global shutdown of both Fable 5 and Mythos 5. The background was a report from Amazon's security researchers about a method of getting around the model's safeguards so that it could find vulnerabilities in software.

26–30 June — the block is lifted

On 26 June the government approved making Mythos 5 available again to a group of U.S. organisations. On 30 June the export controls on both Fable 5 and Mythos 5 were lifted entirely.

1 July — the redeployment

The next day Fable 5 was back for users worldwide, with a new safety classifier that blocks the reported attack method in over 99% of cases and sends those calls on to Opus 4.8. Mythos 5 was still handled separately.

The shutdown is believed to be the first time a leading AI company took a publicly launched model offline after direct intervention from the federal government.

What can Fable 5 do?

• Software development: handles whole codebases autonomously and holds together tasks that run over several days.
• Vision: reads exact numbers from scientific charts and rebuilds web apps directly from screenshots.
• Research: Mythos 5 sped up protein design roughly 10 times, and researchers preferred the model's molecular biology hypotheses in 80% of cases in blind tests.

Fable 5 or Opus 5 — which should you choose?

Choose Fable 5.1 when the task is large, autonomous and expensive to get wrong: a migration across many files, a deep analysis or work that should run for days. Choose Opus 5 or Sonnet 5 for everyday work, texts and simpler code. The price difference is large, so start with the cheaper model and move up only where the result falls short.

What does it mean for businesses and developers?

Nineteen days is a long time if your product rests on a single supplier. A few things worth taking away:

• Do not build yourself into a single model. An abstraction layer that lets you switch supplier costs little now and saves you the day access changes.
• Have a plan B. Which parts of the business stop if your primary model disappears for three weeks?
• Count the cost, not just the capacity. Also count who actually needs the most expensive model.
• Follow the regulations, not just the releases. What a model may do is increasingly governed by legislation and export control, especially for us in Europe.

AI went from tech product to national security issue in three days, and back again in nineteen. The question is no longer just what the models can do, but who gets to use them and how fast that can change. For Swedish and European companies it underlines something we see more and more in our consulting work: whoever understands both the technology and the rules around it stands on firmer ground. If you need help choosing a model and building so you do not lock yourself in, see our AI consulting.

New to how these models work under the hood? Start with what an LLM is, and for OpenAI's equivalent flagship models see our walkthrough of GPT-5.6: Sol, Terra and Luna.