Claude Fable 5 puts Anthropic's capacity trade-offs in public view

Anthropic has released Claude Fable 5, a broadly available Mythos-class model for software engineering, knowledge work, vision, scientific research and longer-running tasks. The company has also introduced Claude Mythos 5 for narrower trusted-access programmes, including Project Glasswing partners and selected biology researchers.

The launch combines a capability announcement with unusually direct capacity management. Fable 5 is available immediately through the Claude API as claude-fable-5. It is also included on Pro, Max, Team and seat-based Enterprise subscription plans at no extra cost until 22 June. From 23 June, Anthropic says subscription-plan access will move to usage credits unless capacity allows the company to extend the included period.

API pricing is $10 per million input tokens and $50 per million output tokens for both Fable 5 and Mythos 5. Anthropic says that is less than half the price of Claude Mythos Preview, although it still places Fable 5 firmly in the premium-model tier. The temporary subscription window gives individual users and teams a short evaluation period before they have to decide where the model earns its cost.

The capacity warning is the important operational detail. Models aimed at long coding sessions, repository-wide migrations, document-heavy analysis and research workflows can create sharp demand spikes. Anthropic has chosen to give subscribers access now, while attaching a date and a usage-credit fallback rather than leaving users to infer the limits from throttling behaviour.

That approach also creates a product challenge. If Fable 5 proves materially better on long software engineering tasks, subscribers may quickly build workflows around it. Anthropic cites early testing from Stripe, where the model reportedly compressed a codebase-wide migration in a 50-million-line Ruby codebase from more than two months of team effort into a day. Even allowing for the usual caution around customer examples, it sets expectations for work that looks closer to engineering infrastructure than casual assistant use.

Fable 5 also arrives with a visible safety distinction. Anthropic says some requests are routed to Claude Opus 4.8 instead, particularly in areas such as cybersecurity. The company says these safeguards trigger in less than 5% of sessions on average, while acknowledging that conservative routing may catch harmless requests. Mythos 5 uses the same underlying model with some safeguards lifted for trusted access programmes.

For developers, the release calls for targeted evaluation rather than a default model switch. The most relevant tests are long, ambiguous, multimodal or expensive in human time: migration work, persistent coding agents, research synthesis, screenshot-to-code tasks and complex document analysis. Routine drafting, classification and first-pass exploration can stay with cheaper models unless Fable 5 shows a clear completion-rate advantage.

The broader point is economic as much as technical. Frontier access increasingly depends on allocation, safety controls and price design as well as raw model quality. Fable 5 gives Anthropic a strong capability story, but it also shows how model providers are beginning to treat their best systems as scarce compute products with managed access rather than unlimited subscription features.

Sources: Anthropic announcement, Claude model overview, Claude pricing.