Claude Fable 5 & Mythos 5: The Complete Guide

Fable 5 and Mythos 5, the first models in Claude's 5th generation: how they differ, plus specs, pricing, safeguards, the Opus 4.8 fallback, and API changes — based on official sources.

Claude Fable 5 and Claude Mythos 5 are the first models in Anthropic's Claude 5 generation, launched on June 9, 2026. Both belong to a new top tier Anthropic calls “Mythos-class” and share the same base model. The biggest difference is availability: Fable 5 is generally available, while Mythos 5 is offered only in limited release.

Claude 5 · Mythos-class Same base model · Launched June 9, 2026 Claude Fable 5 Generally available (API · apps · web) Ships with safety classifiers High-risk requests fall back to Opus 4.8 claude-fable-5 Claude Mythos 5 Limited release (Project Glasswing) No safety classifiers Successor to Mythos Preview claude-mythos-5 Claude Opus 4.8 Fallback model when Fable 5 refuses a request

Anthropic first revealed Claude Mythos Preview in April but limited its rollout because its strong ability to find software vulnerabilities posed misuse risks. Fable 5 is the first model to open this Mythos-class capability to the public safely, by adding safeguards that block responses in high-risk areas.

How Fable 5 and Mythos 5 differ

The two models share the same base model, specs, and pricing. There are two practical differences.

  • Safety classifiers: Fable 5 includes safety classifiers that can decline certain requests. Mythos 5 does not include these classifiers.
  • Availability: Fable 5 is generally available on the Claude API, apps, and web. Mythos 5 is not generally available and is offered only to approved customers in Project Glasswing, a cybersecurity program. Customers without Mythos 5 access can use Fable 5, which offers the same capabilities.

Key specs

ItemClaude Fable 5 / Mythos 5
Model IDclaude-fable-5 / claude-mythos-5
Context window1M tokens by default
Max outputUp to 128k tokens per request
Pricing$10 per 1M input / $50 per 1M output tokens
ThinkingAdaptive thinking, always on
Data retention30-day retention (no ZDR; Covered Model)
Launch dateJune 9, 2026

Both models keep the existing 90% prompt-caching discount on input tokens. US-only inference is available at 1.1x input/output pricing for workloads that must run in the US (per Anthropic).

Safeguards and the Opus 4.8 fallback

The biggest operational change with Fable 5 is that its safety classifiers can decline a request. Questions in high-risk areas such as cybersecurity, biology, and chemistry are not answered by Fable 5; instead they fall back to Claude Opus 4.8, Anthropic's next-most-capable generally available model.

Fable 5 safety classifier & Opus 4.8 fallback User request Claude Fable 5 safety classifier Normal task Fable 5 responds directly High-risk area (cybersecurity · bio · chem) Refusal: stop_reason "refusal" (HTTP 200) → Falls back to Claude Opus 4.8 Requests refused before any output is generated are not billed

API behavior is as follows (per the Claude API docs).

  • Refusal: When Fable 5 declines a request, the Messages API returns stop_reason: "refusal" as a successful HTTP 200 response, not an error, and reports which classifier declined it.
  • Fallback: A refused request can usually be retried on another Claude model in three ways: server-side (the fallbacks parameter, in beta), client-side (SDK middleware), or manual retry.
  • Billing: You are not billed for a request refused before any output is generated. When you retry on another model, fallback credit refunds the prompt-cache cost of switching, so you do not pay it twice.

Performance

Anthropic says Fable 5 exceeds every model it has previously made generally available and is state-of-the-art on nearly all tested benchmarks, with the lead widening on longer, more complex tasks. According to Anthropic's published benchmarks, on SWE-Bench Pro (solving real GitHub issues) it scores about 80%, ahead of Opus 4.8 (about 69%). These figures are Anthropic's own and are not independently verified.

As a real-world example, Anthropic noted that during early testing Stripe ran a codebase-wide migration across a 50-million-line Ruby codebase in a single day — work that would have taken a team over two months by hand.

How to use it

  • Subscribers: Per Anthropic, Fable 5 is included on Pro, Max, Team, and seat-based Enterprise plans at no extra cost from June 9 through June 22, 2026.
  • Developers: Fable 5 is available on the Claude API as claude-fable-5, and is generally available on Claude Platform on AWS, Amazon Bedrock, Vertex AI, and Microsoft Foundry.
  • Supported features: At launch it supports the effort parameter, task budgets (beta), the memory tool, code execution, programmatic tool calling, context editing (beta), compaction, and vision.

API migration checklist

When moving from an earlier model (e.g. Opus 4.8) to Fable 5, check the following (per the Claude API docs).

  • Refusal handling: Add logic for stop_reason: "refusal" (HTTP 200).
  • Fallback path: Configure a server-side, client-side, or manual fallback to retry on another model such as Opus 4.8.
  • Adaptive thinking: It is the only thinking mode and is always on. thinking: {"type": "disabled"} is not supported; control depth with the effort parameter.
  • Thinking output: Raw chain of thought is never returned. Use thinking.display set to "summarized" or the default "omitted", and pass thinking blocks back unchanged on the same model in multi-turn conversations.
  • Data retention: Both models carry 30-day retention and are not available under zero data retention.

Summary

Fable 5 is the most capable model Anthropic has made generally available, opening Mythos-class performance to everyone behind safeguards. Mythos 5 offers the same capability without those safeguards but is limited to approved Project Glasswing customers. Developers should review the refusal/fallback/billing behavior, always-on adaptive thinking, and 30-day retention requirement before migrating.

Sources: Anthropic announcement (anthropic.com/news/claude-fable-5-mythos-5), product page (anthropic.com/claude/fable), Claude API docs (platform.claude.com). Benchmark figures are Anthropic's own.

Keep reading

Have a question or want to share how you use Claude?

Join the community to share tips with other users, or explore more guides.