Claude Fable 5.1 vs Mythos 5.1: What Changed, Pricing and Access

Claude Fable 5.1 and Claude Mythos 5.1 are the same underlying AI model, but they are not the same product. Fable 5.1 is the version most people and developers can use today. Mythos 5.1 applies more permissive safeguards for approved cybersecurity and life-sciences work and is limited to vetted organizations through Anthropic's trusted-access programs.

Anthropic announced both models on September 1, 2026. The practical changes are broader than a benchmark refresh: lower prompt-cache read pricing, new content-provenance measures, more precise cyber safeguards, and several API behaviors developers should test before migrating.

Claude Fable 5.1 vs Mythos 5.1 at a glance

For ordinary Claude users, the decision is simple: use Fable 5.1. Mythos is not a higher-priced public tier that anyone can select. Its purpose is to give verified professionals access to capabilities that standard safeguards may redirect or restrict.

What changed in Claude Fable 5.1?

Anthropic positions Fable 5.1 as a model for demanding coding, knowledge work, research and long-horizon agentic tasks. The company says it improves on Fable 5 while reducing the cost of workloads that reuse cached context.

The model ID in the Claude API is claude-fable-5-1. Anthropic's developer documentation lists a one-million-token context window and a maximum output of 128,000 tokens. Availability includes the Claude API, Amazon Bedrock, Claude Platform on AWS, Google Cloud and Microsoft Foundry. Platform-specific model IDs can differ, so developers should copy the identifier from the documentation for the platform they actually use.

Claude Code defaults to High effort with Fable 5.1, while Claude Cowork and Claude.ai default to Medium effort. That setting matters because higher effort can increase the time and tokens spent on a task. Teams should compare cost per completed task, not only the per-token rate.

Anthropic's benchmark claims need context

In Anthropic's own launch evaluations, Fable 5.1 scored 55.8% on Terminal-Bench 4.0, compared with 42.0% for Fable 5. It also reported 31.4% on AutomationBench versus 17.1% for Fable 5. Mythos 5.1 reached 60.9% on Terminal-Bench 4.0 when evaluated with its different safeguard configuration.

These are vendor-reported results, not a promise that every application will improve by the same amount. The gap between Fable and Mythos can partly reflect safeguard interventions rather than raw reasoning ability. Before switching production traffic, run a representative evaluation set that measures correctness, latency, tool reliability and total cost.

If you rely on AI for research or business decisions, our guide to fact-checking AI answers and citations offers a practical verification workflow that applies to Fable 5.1 as well.

Claude Fable 5.1 pricing explained

Base input and output pricing remains $10 and $50 per million tokens respectively. The major change is prompt-cache reads, which fall to $0.25 per million tokens—a 75% reduction from the previous cache-read price.

Anthropic estimates that token-billed Fable 5.1 workloads will cost about 25% less than Fable 5 in typical use and up to roughly 45% less for highly agentic workflows. Those are estimates, not an automatic discount. The savings depend on repeatedly reusing cached instructions, code, documents or conversation context.

A short, one-off prompt with no cache reuse may see little benefit. A coding agent that repeatedly reads the same repository context, by contrast, can benefit much more. Track cache writes, cache reads and uncached input separately in your usage logs before forecasting savings.

What is Claude Mythos 5.1?

Mythos 5.1 is Fable 5.1 with more permissive safeguards for approved work in two sensitive domains. Anthropic says access will be handled through:

• The Cyber Verification Program: intended for verified defensive-security users. Anthropic says Mythos-class access will be added to the program in the near future.
• The Life Sciences Verification Program: designed for professional research and development. Initial US participants have been enrolled, with broader access planned.

Mythos access should therefore not be described as generally available. Anthropic's developer documentation currently associates it with approved Project Glasswing participants and tells interested organizations to contact their Anthropic, AWS or Google Cloud account team.

The access model resembles a broader industry trend toward separating a public frontier model from a more capable or less restricted version for trusted users. Our coverage of OpenAI Astra's restricted cyber capabilities explains another example of that approach.

Cybersecurity safeguards are less trigger-happy

Fable 5.1 can now help identify software vulnerabilities, a defensive task that earlier safeguards could block too readily. Anthropic says its updated cyber controls produce about 60% fewer interventions per Claude Code session than the safeguards used with Fable 5.

That does not mean unrestricted cyber access. Penetration testing, exploit generation and binary-based vulnerability scanning may still be redirected to other models or refused. Security teams should keep explicit authorization, scope boundaries and human approval gates around any automated testing workflow.

Mythos 5.1 is intended for vetted teams whose legitimate work needs fewer restrictions. Its stronger access is not a substitute for permission to test a system, and organizations remain responsible for their own legal and security controls.

Privacy and Enterprise Frontier Safeguards

Anthropic also announced Enterprise Frontier Safeguards, or EFS. The system is designed to let participating enterprises keep workload data in customer-controlled cloud infrastructure while still supporting automated misuse detection. Anthropic says EFS will roll out in phases beginning in fall 2026.

Eligible customers can receive zero-data-retention access while waiting for EFS, but this is not the default for every Fable 5.1 account. Anthropic's model documentation says Fable 5.1 and Mythos 5.1 otherwise have model-specific retention requirements. Enterprises should confirm their contract and platform configuration rather than assuming that selecting the new model changes retention automatically.

Users of the consumer Claude product should also review their data controls. Our guide to stopping AI services from training on your chats shows where to check those settings across major assistants.

Fable 5.1 adds text watermarking and content provenance

Text produced by Fable 5.1 and Mythos 5.1 carries Anthropic's statistical text watermark across supported platforms. Anthropic introduced the measure for transparency requirements that apply to models released after August 2, 2026 under the EU AI Act.

A watermark-detection API is in private preview for eligible organizations. This does not make every pasted paragraph reliably identifiable by ordinary readers, and a watermark should not be treated as proof that a claim is true. It is a provenance signal, not a fact-checking system.

Supported images, video and audio created through Anthropic's code-execution and Files API workflows can also carry signed C2PA Content Credentials when retrieved through the appropriate API.

Three API changes developers should test before migrating

1. Forced tool use is not supported

Fable 5.1 does not support tool_choice set to any or to a specific named tool. Those requests return a 400 invalid_request_error. Automatic tool selection and disabling tools remain supported.

If an application depends on forcing a particular function call, add an explicit compatibility test before changing its production model ID. A prompt that asks the model to call a tool is not equivalent to an API-level guarantee.

2. Thinking blocks have stricter compatibility rules

Earlier Claude models cannot read Fable 5.1 thinking blocks. Editing earlier messages can also invalidate those blocks. Long-running agents that replay stored transcripts should either start a clean conversation at migration or follow Anthropic's documented rules for preserving compatible context.

3. Anti-distillation controls affect some conversation editing

New API accounts cannot manually alter prior assistant context while preserving the model's earlier thinking transcript. Anthropic says the change will expand gradually and may affect a small number of custom integrations. Test any workflow that rewrites assistant turns, branches conversations or reconstructs tool histories.

Should you upgrade from Fable 5?

Fable 5.1 is worth evaluating if your work involves long coding sessions, repeated repository context, multi-step research, document creation or agents that benefit from prompt caching. The lower cache-read price can make a meaningful difference in those workloads.

Do not migrate solely because a benchmark number is higher. A safe rollout looks like this:

1. Build a test set from real tasks and known-good answers.
2. Run Fable 5 and Fable 5.1 with the effort settings you would use in production.
3. Measure success rate, latency, output tokens, cache reuse and total cost per completed task.
4. Test tool-calling paths, especially if you previously forced tool selection.
5. Review refusals and safeguard interventions in authorized security or scientific workflows.
6. Roll out gradually and preserve a fallback model until results are stable.

Teams connecting Claude to laboratory equipment should also read our explanation of Anthropic's Model Hardware Standard, which addresses how AI systems can interact with physical research instruments under controlled permissions.

The bottom line

Fable 5.1 is the public choice; Mythos 5.1 is the vetted-access choice. They share the same core capabilities and pricing, but their safeguards and availability differ. Most developers should test Fable 5.1, paying special attention to prompt caching, effort settings and the changed tool-selection behavior.

Cybersecurity and life-sciences organizations that need Mythos should apply through the relevant trusted-access channel and wait for approval. No public pricing tier or model switch bypasses that verification.

For full technical details, see Anthropic's launch announcement, the Fable 5.1 model overview and the developer migration notes.