Claude Fable 5: What Anthropic's Public Mythos-Class Model Changes for AI Workflows
Anthropic has released Claude Fable 5, its first generally available Mythos-class model, while keeping Claude Mythos 5 limited to approved partners. The release matters for developers because it combines long-horizon coding and knowledge-work gains with new safety fallbacks, higher pricing, and mandatory 30-day retention.
Claude Fable 5 in brief
Claude Fable 5 is Anthropic's first generally available Mythos-class Claude model. It gives developers and paid Claude users access to the same broad capability tier as Claude Mythos 5, but with safety classifiers that can refuse or route high-risk requests away from the unrestricted model path. For teams building AI workflows, the release is less about a new chatbot name and more about a new tradeoff: stronger long-horizon work, higher cost, and stricter data retention.
Key takeaways
- Claude Fable 5 is generally available through the Claude API and major cloud platforms; Claude Mythos 5 remains limited to approved Project Glasswing customers.
- Anthropic positions Fable 5 as stronger than its previous generally available Claude models for software engineering, knowledge work, vision, science, and long-context tasks.
- The Claude API model ID is
claude-fable-5; Claude Mythos 5 usesclaude-mythos-5but is not broadly available. - Both models are listed at $10 per million input tokens and $50 per million output tokens.
- Claude Fable 5 requires 30-day traffic retention for safety monitoring and is not available under zero data retention terms.
What Fable 5 actually is
Anthropic describes Fable 5 as a Mythos-class model made safe for general use. In Anthropic's own framing, Mythos-class sits above Opus-class capability, with Fable and Mythos separated by safety controls rather than by a completely different public product line.
The practical split looks like this:
| Model | Best use | Access | Main limitation | Source |
|---|---|---|---|---|
| Claude Fable 5 | Public high-end coding, research, long-context work, and agentic workflows | Generally available via API and cloud partners | Safety classifiers, possible refusals or fallback handling, 30-day retention | Anthropic and Claude docs |
| Claude Mythos 5 | Restricted cyber, infrastructure, and future specialist research work | Limited Project Glasswing access | Not generally available | Anthropic and Claude docs |
| Claude Opus 4.8 | Safer fallback and current Opus-tier workflows | Existing Claude surfaces | Lower capability tier than Fable 5 for the hardest tasks | Anthropic and GitHub |
The most important implementation detail is that Fable 5 is not just a "faster Claude" upgrade. It introduces model-specific behavior around refusals, fallback, thinking controls, billing, and retention. Builders should treat it as a new deployment target, not a silent replacement.
Availability and pricing
Anthropic says Claude Fable 5 is available from June 9, 2026 through the Claude API, Claude Platform on AWS, Amazon Bedrock, Google Vertex AI, and Microsoft Foundry. The API model ID is claude-fable-5.
The Claude API docs list a 1 million token context window by default and up to 128,000 output tokens per request for Fable 5 and Mythos 5. Pricing is $10 per million input tokens and $50 per million output tokens.
For subscriptions, Anthropic says Fable 5 is included on Pro, Max, Team, and seat-based Enterprise plans through June 22, 2026 at no extra cost. From June 23, usage credits are required unless capacity allows Anthropic to extend the included window.
Practical LinkLoot angle
If you run AI workflows, Fable 5 is most interesting where model quality can reduce iteration loops. That means multi-file coding tasks, research synthesis, long-context analysis, tool-heavy agents, and review workflows where fewer retries may justify a higher token price.
The first workflow to test is not "replace every Claude call." A safer rollout is:
- Move the hardest 5 to 10 percent of tasks to
claude-fable-5. - Keep cheaper models for short drafting, extraction, classification, and routine chat.
- Add refusal and fallback handling before moving production jobs.
- Re-check privacy terms if you rely on zero data retention today.
For more workflow ideas, see LinkLoot's guide to AI workflow automation.
Data retention and fallback caveats
The biggest non-benchmark caveat is retention. Anthropic says Mythos-class traffic requires 30-day retention for safety monitoring and that retained traffic is not used to train new Claude models. GitHub's Copilot changelog mirrors the same point: Claude Fable 5 is off by default for Copilot Business and Enterprise administrators because it requires prompt and output retention for safety classifiers.
The second caveat is fallback behavior. Anthropic says Fable 5 can hand off certain high-risk categories, including cybersecurity, biology, chemistry, and distillation-related requests, instead of serving them through the full Fable path. Early Anthropic data says more than 95% of sessions do not require fallback, but production teams still need to handle refusal and fallback responses explicitly.
What developers should check first
Before moving a production agent or coding workflow to Fable 5, verify these details:
- API behavior: the Messages API can return a successful HTTP response with a refusal stop reason, so refusal is not necessarily an HTTP error.
- Budgeting: the higher token price may be offset by fewer retries, but only if you measure complete task cost rather than price per token alone.
- Platform access: GitHub Copilot is rolling out Fable 5 gradually and requires admins to enable the model policy for Business and Enterprise plans.
- Privacy posture: Fable 5 and Mythos 5 are covered models with 30-day retention, not zero data retention.
- Prompt migration: the Claude docs say Fable 5 uses the same broad prompting techniques, but long-context and reasoning instructions need careful structure.
Source check
Anthropic's announcement confirms the release, the Fable/Mythos split, the safety classifier rationale, the fallback categories, the $10/$50 per million token pricing, the staged subscription access window, and the 30-day retention policy.
The Claude API docs confirm model IDs, availability surfaces, context/output limits, refusal behavior, fallback guidance, billing notes, supported features, and the zero-data-retention caveat.
GitHub's changelog confirms Copilot availability, gradual rollout, admin policy requirements, usage-based billing, and the same 30-day retention requirement for Copilot customers.
TechCrunch and All-AI independently summarize the public release, the restricted Mythos 5 program, the safety guardrails, and the practical enterprise concerns around cost and retention.
Yes. Anthropic says Claude Fable 5 is generally available through the Claude API and major cloud platforms from June 9, 2026.
Bottom line
Claude Fable 5 is a major release for developers who need stronger long-horizon reasoning, coding, and tool use, but it is not a low-friction default upgrade. The teams that benefit most will route only high-value tasks to Fable 5, measure end-to-end cost, and treat retention plus refusal handling as launch requirements rather than footnotes.