Anthropic launches Claude Fable 5

Anthropic's automatic fallback routing to an older model for high-risk prompts is a clever compromise for safety, but it essentially forces developers to pay premium rates while potentially receiving downgraded outputs when working on complex or dual-use tasks.

• –**Dual-Tier Model Access:** Splitting the release into the public-facing, heavily guarded Claude Fable 5 and the private, unrestricted Claude Mythos 5 (via Project Glasswing) demonstrates how frontier AI labs are managing liability while keeping power users happy.
• –**Automated Fallback Compromise:** Routing restricted prompts to Claude Opus 4.8 keeps the user session alive, but highlights the friction of running real-world security or biology pipelines that trip safety classifiers.
• –**High-Horizon Autonomy:** With a 128,000 token output limit, Fable 5 is optimized for large-scale code synthesis and multi-step plans, signaling that Anthropic is betting heavily on agentic workflows rather than simple chat.