Anthropic Launches Claude Fable 5, Its First Public Mythos-Class Model
On June 9, Anthropic released Claude Fable 5, the first publicly available model in a new capability tier the company calls Mythos. For a roughly two-week window, Fable 5 is bundled into paid Pro, Max, Team, and Enterprise plans at no extra cost before moving to usage-based pricing on June 23. The model posts record benchmark scores, but sensitive prompts are silently routed to Claude Opus 4.8 by additional safety classifiers, and all business traffic carries a mandatory 30-day data-retention window.
What Mythos Is, and How Fable 5 Differs From Mythos 5
Anthropic organizes its models into capability tiers: Haiku, Sonnet, Opus, and now Mythos. Fable 5 and its restricted sibling, Mythos 5, are the same underlying model. The difference is the guardrails. Anthropic determined that Mythos-level capability in cybersecurity and biology had crossed a threshold where it could meaningfully assist bad actors, so the public version ships with an additional layer of safety classifiers.
The mechanic behind those classifiers is what makes the launch unusual. When a prompt touches a sensitive domain (cybersecurity, biology, chemistry) or attempts to mass-extract the model’s behavior, a classifier intercepts it and routes the response to Claude Opus 4.8 instead of Fable 5. The user still receives a full answer, just from the older model, and is notified of the swap. Anthropic says this affects under 5 percent of sessions, so most users interact with full-strength Fable 5 the vast majority of the time.
How Does Fable 5 Perform on Benchmarks?
According to Anthropic’s launch announcement, Fable 5 posts record results across most of the benchmarks the company tests, with its lead widening on longer and more complex tasks:
- 80.3 percent on SWE-bench Pro, a hard software-engineering benchmark, against 58.6 percent for the competing frontier model.
- First model past 90 percent on one analytics platform’s test of senior-analyst-level tasks, roughly 10 points above Opus.
- An early enterprise tester reported using it to migrate a 50-million-line codebase in a single day, work it estimated would have taken a human team more than two months.
The vision results are among the more practically interesting. Anthropic highlights that Fable 5 can extract accurate numbers from charts and reconstruct a web application’s layout and styling from a screenshot alone, the difference between describing an interface you want built and simply handing the model a reference image.
What Does Fable 5 Cost, and When Does the Free Window Close?
Through the API, Fable 5 is priced at $10 per million input tokens and $50 per million output tokens, which Anthropic notes is more than twice as cheap as its earlier Mythos preview but still well above Opus-tier pricing. For subscribers, free access is the headline, and it is brief. From June 9 through June 22, Fable 5 is included in paid plans at no additional charge. Starting June 23, it moves to usage credits, meaning subscribers pay per use until Anthropic has enough compute to fold it back into standard plans. The company has been explicit that demand is expected to be heavy and the rollout will be phased.
The Data-Retention Catch for Businesses
There is a meaningful condition for organizations. All business traffic on Mythos-class models carries a mandatory 30-day data-retention window that cannot be disabled. Anthropic says it will not train on that data and will delete it after 30 days, with logged employee access. But for teams operating under strict client-confidentiality or regulatory requirements, an inability to opt out of retention is a detail worth clearing with legal or compliance before routing sensitive material through the model. Fable 5 is also not available under zero-data-retention configurations at all.
The Real Innovation Is the Release Strategy
The benchmark numbers are notable, but the more durable development is the playbook. Anthropic is shipping its most capable system to the public by selectively dialing specific capabilities down, routing sensitive prompts to a quieter, less capable fallback rather than issuing outright refusals, after first proving the model out with vetted partners. It is a staged path: lock the capability down, validate it with trusted users, then loosen access while keeping a safety net in place.
That approach is likely how frontier AI continues to arrive: restricted first, broadened later, with capability gated by domain rather than withheld entirely. For anyone evaluating AI tools, the practical implication is that "the most capable model" is becoming a moving and conditional target. The version you can access, the price you pay for it, and the guardrails wrapped around it can all change within weeks of a launch.
What Should Teams Do Before June 23?
The two-week free window is effectively a no-cost evaluation period for a frontier model, and it closes fast. Teams that want to know whether Mythos-class capability is worth the premium over Opus 4.8 have a narrow opportunity to test it against real workloads before June 23. After that, the decision becomes a per-token cost question, and the answer will depend on whether the measurable gains on long, complex, vision-heavy tasks justify the price for the specific work at hand.
FAQ
What is the difference between Claude Fable 5 and Mythos 5?
Fable 5 and the restricted Mythos 5 are the same underlying model. The public Fable 5 ships with extra safety classifiers that intercept sensitive prompts in cybersecurity, biology, and chemistry (and mass-extraction attempts) and reroute them to Claude Opus 4.8, a swap that Anthropic says affects under 5 percent of sessions.
How much does Claude Fable 5 cost, and when is the free access window?
API pricing is $10 per million input tokens and $50 per million output tokens. From June 9 through June 22, Fable 5 is included at no extra cost in Pro, Max, Team, and Enterprise plans. Starting June 23, it moves to usage credits, meaning subscribers pay per use until Anthropic has enough compute to fold it back into standard plans.
What is the data-retention policy for Claude Fable 5 in business plans?
All business traffic on Mythos-class models carries a mandatory 30-day data-retention window that cannot be disabled, and Fable 5 is not available under zero-data-retention configurations. Anthropic states it will not train on that data and will delete it after 30 days, with logged employee access.