Anthropic’s Two-Tier Model Deployment: Fable 5, Mythos 5, and the 19-Day Export Control Crisis

• Fable 5 includes runtime safety classifiers that intercept requests in sensitive domains (cybersecurity, biology, chemistry, model distillation) and downgrade them in-flight to Claude Opus 4.8 . The classifiers trigger on average in less than 5% of sessions .
• Mythos 5 removes those classifiers, allowing unrestricted inference, and is gated to a small, vetted group of cybersecurity defenders and critical-infrastructure partners .

Context Window, Output, and Pricing

Both models share identical specs and pricing:

• Context window: 1 million tokens (default)
• Maximum output: Up to 128k tokens
• Pricing: $10 per million input tokens, $50 per million output tokens
• Prompt caching discount of 90% on input tokens is available
• US-only inference available at 1.1x base pricing

Export-Control Ban and Rollback Timeline

• June 12: The US government applied export controls citing national security, ordering Anthropic to block access to Fable 5 and Mythos 5 for any foreign national . Because Anthropic had no reliable way to verify nationality in real-time, it disabled both models for all users worldwide by June 13 .
• June 26: Commerce Secretary Howard Lutnick partially lifted the directive — re-enabling Mythos 5 for over 100 vetted US critical-infrastructure organizations, while Fable 5 remained offline globally .
• June 30: The US Department of Commerce fully lifted export controls on both Fable 5 and Mythos 5 .
• July 1: Anthropic reinstated Fable 5 globally across Claude.ai, the Claude Platform, AWS, Microsoft Foundry, and Claude Code .

MixRoute's Single-Key Access (July 4)

On July 4, 2026, MixRoute — a Singapore-based unified API gateway supporting 200+ models with a single API key — confirmed it will offer OpenAI's GPT-5.6 model family as soon as the models reach general availability, alongside Claude Fable 5, which is already live on the platform . GPT-5.6 remains in a limited preview gated to a small, government-vetted group of partners and has not yet reached general availability .

Broader Industry Implications

Shipping identical model weights under two governance regimes represents a structural shift in frontier AI deployment:

• Safety vs. capability balancing: Fable 5's safety-routing classifier (falling back to Opus 4.8 for high-risk queries) creates a practical "speed bump" rather than a capability ceiling, allowing public access to frontier intelligence while reserving unfiltered capability for trusted entities .
• Export control precedent: The rapid imposition and reversal of export controls on a software model (not hardware) established a new regulatory flashpoint — the US government demonstrated willingness to cut global access to frontier models, but reversed course within three weeks under industry pressure .
• RTO (Responsible/Tiered Operation) model: Other labs may adopt similar dual-SKU deployments — a "public guarded" variant and a "trusted-partner unrestricted" variant — to navigate the tension between publishing frontier capability and satisfying government non-proliferation demands .