claudekit / updates / claude-fable-5-mythos-5
[ NEW · ]

Claude Fable 5 and Claude Mythos 5

Anthropic released Claude Fable 5 and Claude Mythos 5 on June 9, 2026. Fable 5 is a Mythos-class model made safe for general use — state-of-the-art on nearly all tested benchmarks, with the `claude-fable-5` model ID. Mythos 5 is identical with cybersecurity safeguards lifted, available in limited release through Project Glasswing. Both support a 1M token context window and up to 128k output tokens by default, priced at $10/$50 per MTok. When a safety classifier declines a request, the Messages API returns stop_reason: "refusal" as an HTTP 200, and the fallbacks parameter (beta) or SDK middleware retries on another Claude model.

ModelAPI
Official announcement →

This article is a summary based on official documentation.

Overview

Anthropic released Claude Fable 5 and Claude Mythos 5 on June 9, 2026. They share the same underlying model: Fable 5 is the Mythos-class model made safe for general use with safeguards enabled, and Mythos 5 lifts the cybersecurity safeguards for approved Project Glasswing customers. Fable 5 is state-of-the-art on nearly all tested benchmarks, with the claude-fable-5 model ID and $10/$50 per MTok pricing.

Key features

  • Mythos-class capability, generally available

    Mythos-class capability was previously limited to select customers as Claude Mythos Preview. Fable 5 brings the same class of model to general availability with three classifier systems (cybersecurity, biology/chemistry, distillation) enabled, and excels at longer, more complex tasks. Stripe reported Fable 5 compressed months of work into days, performing a 50M-line Ruby migration in one day versus an estimated two months of manual effort.

  • 1M token context window and long-horizon focus

    Fable 5 maintains focus across millions of tokens. Both models support a 1M token context window and up to 128k output tokens per request by default, and Fable 5 improved Slay the Spire performance 3x over Opus 4.8.

  • State-of-the-art vision

    Fable 5 is state-of-the-art for vision tasks and completed Pokémon FireRed using a vision-only approach.

  • Adaptive thinking only, controlled by effort

    Adaptive thinking is the only thinking mode on Claude Fable 5 and Claude Mythos 5. It applies whenever the thinking parameter is unset, and thinking: {"type": "disabled"} is not supported — use the effort parameter to control thinking depth. The raw chain of thought is never returned: thinking.display defaults to "omitted" (thinking blocks with an empty thinking field), and display: "summarized" returns readable summarized thinking.

  • Refusals and fallback

    When a safety classifier declines a request, the Messages API returns stop_reason: "refusal" as a successful HTTP 200 response — not an error — and reports which classifier declined. Pass the fallbacks parameter (beta on the Claude API and Claude Platform on AWS) to have the API retry on another Claude model, or use the SDK middleware (TypeScript, Python, Go, Java, and C#) to retry from the client on any platform. You are not billed for a request refused before any output is generated, and fallback credit refunds the prompt-cache cost of switching models on retry.

  • Mythos 5 through Project Glasswing

    Mythos 5 shares Fable 5’s capabilities with the cybersecurity safeguards lifted. It is restricted to cyberdefenders and infrastructure providers, deployed through Project Glasswing in collaboration with the US government, with biology/chemistry safeguards lifted for select biology researchers. It succeeds Claude Mythos Preview. In drug design, Mythos 5 accelerated a protein design process roughly 10x and identified 9 of 14 strong drug candidates.

Benchmark performance

AreaResult
Hebbia Finance BenchmarkHighest score for senior-level reasoning
Cognition FrontierCodeTop performer
Core analytics benchmarkFirst to break 90% (10-point jump over Opus 4.8)
ViBench (end-to-end vibe-coding)Highest score

Pricing and availability

ItemDetail
Input / output$10 / $50 per MTok (less than half the price of Claude Mythos Preview)
Model IDsclaude-fable-5, claude-mythos-5
Context window1M tokens by default
Max output128k tokens
Platforms (Fable 5)Claude API, Claude Platform on AWS, Amazon Bedrock, Vertex AI, Microsoft Foundry
Subscription plansIncluded on Pro, Max, Team, Enterprise June 9-22; usage credits required from June 23
Mythos 5Approved Project Glasswing customers only (contact your Anthropic, AWS, or Google Cloud account team)

Notes

  • Safeguards trigger on less than 5% of sessions on average — the cybersecurity classifier blocks offensive cyber tasks (zero compliance with harmful single-turn cyberattack requests in external testing), most biology/chemistry requests conservatively fall back to Opus 4.8, and the distillation classifier blocks attempts to extract model capabilities for unauthorized uses.
  • Covered Models data retention — both models carry 30-day data retention and are not available under zero data retention. Retained data is not used for training new models or non-safety purposes, and human access is logged.
  • Pass thinking blocks back unchanged — in multi-turn conversations, return thinking blocks unchanged on the same model. See the docs on thinking output for cross-model handling.
  • Supported features at launch — effort, task budgets (beta: task-budgets-2026-03-13 header), the memory tool, tool result clearing through context editing (beta: context-management-2025-06-27 header), compaction, and vision.
  • Migration guides — step-by-step instructions are available for migrating from Claude Mythos Preview and from Claude Opus 4.8 on platform.claude.com.
  • Alignment assessment — Mythos 5 showed low levels of misaligned behavior comparable to Opus 4.8, and similar alignment is expected for Fable 5 given the identical underlying model.
§ 6

Frequently Asked Questions

frequently asked
§ 6.1
What is Claude Fable 5?
Anthropic's most capable widely released model — a Mythos-class model made safe for general use with safeguards enabled. It is state-of-the-art on nearly all tested benchmarks and excels at longer, more complex tasks. The API model ID is `claude-fable-5`.
§ 6.2
What's the difference between Fable 5 and Mythos 5?
They share the same underlying model. Fable 5 ships with three classifier systems (cybersecurity, biology/chemistry, distillation) enabled and is generally available. Mythos 5 lifts the cybersecurity safeguards and is restricted to approved Project Glasswing customers — cyberdefenders and infrastructure providers.
§ 6.3
How much does it cost?
$10 per million input tokens and $50 per million output tokens — less than half the price of Claude Mythos Preview. You are not billed for a request that is refused before any output is generated.
§ 6.4
Where and when is it available?
Both models became available on June 9, 2026. Fable 5 is generally available on the Claude API, Claude Platform on AWS, Amazon Bedrock, Vertex AI, and Microsoft Foundry. On subscription plans it is included on Pro, Max, Team, and Enterprise from June 9-22, and requires usage credits from June 23.
§ 6.5
What happens when a safety classifier declines a request?
The Messages API returns stop_reason: "refusal" as a successful HTTP 200 response — not an error — and reports which classifier declined. Pass the fallbacks parameter (beta) to have the API retry on another Claude model, or use the SDK middleware (TypeScript, Python, Go, Java, C#) to retry client-side.
§ 6.6
Does it affect existing Opus 4.8 code?
Adaptive thinking is the only thinking mode — thinking: {"type": "disabled"} is not supported — and raw thinking content is never returned (thinking.display defaults to "omitted"). The Messages API is unchanged for Opus, Sonnet, and Haiku. See the official migration guide when moving from Opus 4.8.