Anthropic releases Claude Fable 5, its first publicly available Mythos-class model

Anthropic released Claude Fable 5 on June 9, 2026, becoming the first publicly available model from its Mythos-class family. According to The Verge, the model demonstrates “exceptional performance in software engineering, knowledge work, and vision,” with capability gains accelerating as task complexity increases. The release follows months of internal debate over whether Mythos-class capabilities in cybersecurity posed too great a risk for unrestricted distribution.

Safeguards and the Fallback Architecture

The key mechanism enabling Fable 5’s public release is a layered safety system that blocks responses in designated high-risk domains. According to The Verge, Anthropic identified cybersecurity and biology as the primary areas where safeguards trigger, causing the model to defer to Claude Opus 4.8—Anthropic’s previously released flagship model—rather than refusing outright. In internal testing, Anthropic reports that 95 percent of Fable 5 sessions completed entirely using Fable responses without triggering a fallback to Opus 4.8.

This architecture represents a middle path between unrestricted release and continued gating. Rather than withholding the model entirely, Anthropic segments responses by domain sensitivity, allowing Fable 5 to operate freely on non-sensitive tasks while maintaining human-reviewable constraints on high-risk domains.

Claude Mythos 5 and Tiered Access

Anthropic simultaneously announced Claude Mythos 5, described as the same underlying model as Fable 5 but “with the safeguards lifted in some areas.” According to The Verge, access to Mythos 5 appears limited to organizations enrolled in Anthropic’s private Project Glasswing initiative, which grants early access to high-capability model variants. The company indicated plans to “expand access over time through a more systematic trusted-access program,” though specific timelines and selection criteria remain undisclosed.

The Verge notes that Anthropic did not provide on-the-record comment on why the models are numbered “5” despite no previously released Mythos or Fable models, leaving the version-numbering convention unexplained.

Fable 5 Pricing and Market Position

Claude Fable 5 carries significantly higher per-token costs than Anthropic’s prior public flagship. According to Anthropic, pricing is set at $10 per million input tokens and $50 per million output tokens—double the rate of Claude Opus 4.8. This pricing places Fable 5 in a premium tier, reflecting its enhanced capabilities across longer and more complex reasoning tasks.

Why This Matters

The Fable 5 release signals Anthropic’s confidence in domain-specific safety mechanisms as a viable alternative to full model gating. Organizations requiring unrestricted Mythos-class performance in sensitive domains now have a clear incentive to seek Project Glasswing access—potentially expanding Anthropic’s footprint in regulated industries like financial services and healthcare. For teams already using Claude Opus 4.8 on cybersecurity or biology tasks, Fable 5’s fallback architecture preserves safety guarantees while enabling migration to a more capable baseline for non-sensitive work. The tiered pricing structure and staged access rollout suggest Anthropic is using Mythos as a test bed for more granular capability and trust tiers, a pattern likely to influence how frontier labs manage capabilities-vs.-safety trade-offs as model capabilities continue to expand.