Claude Fable 5 Quietly Downgrades to Weaker Model Mid-Session

Anthropic’s real news here is not that a powerful model launched. It’s that the company is experimenting with automatic capability downgrading mid-session, quietly routing users to a weaker model without explicit notice.

Claude Fable 5 is Anthropic’s first publicly available model from its Mythos class, a tier previously kept behind closed doors for partners and cybersecurity organizations. The company positions it as a step forward in autonomous, long-running tasks, including software engineering, extended research, and complex multi-step workflows where the AI handles context across longer sessions without constant user prompting.

The safety mechanism buried in the announcement deserves more attention than it’s getting. When Fable 5 detects certain sensitive requests, particularly in cybersecurity, it automatically falls back to Claude Opus 4.8, a less capable model, without necessarily alerting the user. Anthropic says this affects fewer than 5% of sessions on average, and acknowledges that some harmless requests get caught in the filter.

That 5% figure sounds small until you consider it at scale across enterprise deployments, where a silent capability downgrade mid-workflow could produce inconsistent outputs that are difficult to trace.

Pricing and Access

Both Fable 5 and the more restricted Claude Mythos 5 are priced at $10 per million input tokens and $50 per million output tokens. That’s less than half the cost of the earlier Mythos Preview, which is a meaningful reduction for teams running high-volume workloads.

Access has a catch. Fable 5 is free on Pro, Max, Team, and Enterprise plans through June 22, after which usage credits are required. Anthropic says broader subscription access will return once capacity allows, which is a softer commitment than it sounds and leaves the timeline entirely open.