Anthropic updated its platform documentation for Claude Fable 5 and Claude Mythos 5 on July 19, with expanded guidance on agentic integration patterns including refusal handling, automated model fallback, and adaptive thinking configuration, according to the updated documentation page.

The documentation positions Fable 5 as “Anthropic’s most capable widely released model, built for the most demanding reasoning and long-horizon agentic work.” Mythos 5 shares the same capabilities but ships without Fable 5’s safety classifiers, available only through the restricted Project Glasswing program.

Agent-Specific Integration Changes

The documentation highlights three integration changes specific to agent builders:

Refusal handling. Fable 5 includes safety classifiers that can decline requests. When this happens, the API returns stop_reason: "refusal" as a successful HTTP 200 response rather than an error. Agent builders need to handle this new response type, which includes which classifier triggered the refusal. Mythos 5 does not include these classifiers.

Automated fallback. A refused Fable 5 request can be retried on another Claude model through three mechanisms: server-side (via a fallbacks API parameter, currently in beta), client-side (via SDK middleware), or manual retry logic. Billing credits refund the prompt-cache cost of switching models, so agents that hit a refusal and retry do not pay the caching cost twice.

Adaptive thinking. Adaptive thinking is the only thinking mode on both models. It cannot be disabled. Raw chain-of-thought content is never returned. Instead, agents receive either a summarized reasoning block or an empty thinking field, controlled by a thinking.display setting. This represents a departure from Opus-era models where thinking could be toggled on or off.

Specs and Pricing

Both models share the same specifications: a 1 million token context window by default, up to 128,000 output tokens per request, and pricing at $10 per million input tokens and $50 per million output tokens. Both carry 30-day data retention and are not available under zero data retention, which may affect enterprise deployments with strict data governance requirements.

Tiered Access Strategy

The documentation update follows Anthropic’s July 19 announcement that Claude Fable 5 is now available to Claude Max and Team Premium subscribers at 50% of usage limits, shifting from temporary to permanent tiered access. Together, the moves outline a two-layer monetization approach: general availability through API with usage-based pricing, and consumer subscription tiers with capped access.

Mythos 5 remains the most restricted model in Anthropic’s lineup. It is offered only to approved Project Glasswing customers through Anthropic, AWS, or Google Cloud account teams. The documentation notes that “customers without access to Claude Mythos 5 can use Claude Fable 5, which is generally available and offers the same capabilities,” drawing the distinction purely in terms of safety classifiers rather than raw capability.

Both models became available on June 9, 2026, across the Claude API, Amazon Bedrock, Claude Platform on AWS, Google Cloud, and Microsoft Foundry.