Anthropic now offers two June 2026 frontier tiers with the same intelligence but different safety envelopes: Claude Fable 5 (GA) and Claude Mythos 5 (Glasswing).
Identical capabilities
1M context, 128K output, adaptive thinking, same API pricing, same tool support (memory, code execution, programmatic tool calling, vision).
The only difference: refusals
Fable 5 can decline requests via safety classifiers. Mythos 5 has no classifiers — completions always attempt an answer. Mythos is for controlled enterprise environments with external guardrails.
Access
Fable 5: everyone. Mythos 5: Project Glasswing approval only. No public signup.
Recommendation
Default to Fable 5 with fallback engineering. Request Mythos 5 only if refusals break a validated internal pipeline and your compliance team approves reduced classifier coverage.
