What's New

Anthropic's Claude Fable 5, the first generally available Mythos-class model, launched on AWS on June 9, 2026, bringing frontier-level autonomous reasoning and coding capabilities to all customers via Amazon Bedrock and Claude Platform on AWS. A restricted variant, Claude Mythos 5 (without safety classifiers), was simultaneously made available to a limited set of pre-approved customers. However, on June 12, 2026, the US Government issued an export control directive requiring Anthropic to revoke access to both Fable 5 and Mythos 5 for all users, citing national security concerns related to a reported jailbreak technique.

How It Works

• Autonomous long-horizon operation: Fable 5 can execute complex, multi-day knowledge work and coding tasks without human intervention, planning across stages, delegating to sub-agents, and self-verifying outputs before delivery.
• Adaptive reasoning: Extended thinking (adaptive mode) is always enabled and cannot be disabled; reasoning effort level is configurable by the caller to balance latency and depth.
• Large context window: Supports a 1M-token context window with up to 128K output tokens per request, enabling ingestion of very large codebases, documents, or datasets in a single call.
• Safety classifiers (Fable vs. Mythos): Fable 5 includes blocking classifiers for dual-use cybersecurity and biology content; Mythos 5 is the same underlying model without those classifiers, available only to vetted customers.
• Dual access paths: Amazon Bedrock (AWS-operated inference, data stays in AWS, integrates with Guardrails/Knowledge Bases/regional residency) and Claude Platform on AWS (Anthropic-operated infrastructure, same-day feature access, unified AWS billing via Marketplace).
• Prompt caching: Supported with a minimum of 1,024 tokens per checkpoint, up to 4 checkpoints per request, and TTLs of 5 minutes or 1 hour on system, messages, and tools fields.
• Constrained sampling: Temperature must be 1.0 or unset; top_p must be ≥ 0.99 and < 1.0 or unset; top_k is not supported — reflecting the model's design for high-fidelity, deterministic-leaning outputs.
• Refusal handling: Content blocked by classifiers returns HTTP 200 with stop_reason: "refusal" and a stop_details object; prompt-stage refusals are not billed, but mid-stream refusals are billed for tokens generated before the block.
• Data retention requirement: Fable 5 mandates 30-day customer data retention on Amazon Bedrock to support Anthropic's monitoring and rapid-response strategy against jailbreak attempts.

Why It's Important

• Democratization of frontier AI: Fable 5 was the first time Mythos-class (frontier reasoning) capabilities were made broadly available rather than restricted to a small preview cohort, representing a significant lowering of the barrier to production-grade autonomous AI.
• Step-change in autonomous work: The ability to run unsupervised for multi-day tasks across finance, legal, engineering, and data domains signals a qualitative shift from AI as a copilot to AI as an autonomous worker.
• Export control precedent: The June 12 suspension under a US Government export control directive is a landmark event — the first known instance of a frontier AI model being pulled from a major cloud platform for national security reasons, with broad implications for AI governance and enterprise risk planning.
• Defense-in-depth safety model: Anthropic's published strategy of layered safeguards, mandatory data retention, and rapid-response monitoring sets a new industry reference point for how to deploy powerful models responsibly at scale.
• Enterprise workflow integration: Native support for AWS Guardrails, Knowledge Bases, IAM, and unified billing lowers integration friction for enterprises already operating within the AWS ecosystem.

How It's Different

• Mythos-class capabilities with guardrails: Unlike prior generally available Claude models, Fable 5 brings the same underlying capability as the restricted Mythos 5 but adds production-grade safety classifiers, making frontier reasoning accessible without requiring special vetting.
• Always-on adaptive thinking: Unlike models where extended reasoning is an optional toggle, Fable 5's adaptive thinking is permanently enabled, ensuring the model always applies its full reasoning capacity rather than defaulting to shallow responses.
• Self-updating and self-verifying: Fable 5 proactively updates its own skill representations based on task learnings and builds its own evaluation harnesses — behaviors not present in prior Claude generations.
• Two distinct deployment models on AWS: The Bedrock path (AWS-operated, data residency guarantees) and Claude Platform on AWS (Anthropic-operated, same-day feature parity) give customers an explicit choice between compliance-first and feature-first deployment, unlike most model integrations that offer only one path.
• Stricter sampling constraints: The locked temperature and top_p range distinguish Fable 5 from earlier Claude models, reflecting a deliberate design choice to optimize for reliable, high-quality autonomous outputs over creative variability.
• Mandatory 30-day data retention: This is a new policy requirement not present on other Claude models, tied directly to Anthropic's security monitoring obligations for a model with this capability level.

When to Prefer It

• Long-horizon autonomous coding: When you need an agent to independently plan, implement, test, and iterate on a complex software project over hours or days without continuous human prompting.
• Professional knowledge work at scale: When automating high-complexity tasks in finance (e.g., financial modeling, regulatory analysis), legal (e.g., contract review pipelines), or data engineering where multi-step reasoning and self-verification are critical.
• AWS-native compliance environments: When your organization requires data to remain within a specific AWS region or geography and you need a frontier model that integrates natively with Guardrails, Knowledge Bases, and IAM.
• Enterprises needing unified billing: When you want to consolidate AI spend on your existing AWS invoice rather than managing a separate Anthropic billing relationship.
• Feature-parity with Anthropic's first-party API: When you need same-day access to new Claude features, Agent Skills, code execution, or Files API — use Claude Platform on AWS rather than Bedrock.
• High-throughput batch workloads: When processing large volumes of complex documents or code artifacts where the 1M-token context window and batch processing support provide a meaningful throughput advantage.
• Note — currently not applicable: As of June 12, 2026, access has been suspended for all users pending resolution of the US Government export control directive; customers should monitor Anthropic's official communications before planning new deployments on Fable 5.

Availability

• General availability status: Claude Fable 5 launched as generally available on Amazon Bedrock and Claude Platform on AWS on June 9, 2026.
• Access suspended: Effective June 12, 2026, access to both Claude Fable 5 and Claude Mythos 5 has been revoked for all users on Amazon Bedrock and Claude Platform on AWS pursuant to a US Government export control directive; no timeline for reinstatement has been provided.
• Unaffected models: All other Amazon Bedrock models, including Claude Opus 4.8, remain fully available and are not subject to the directive.
• Mythos 5 access: Claude Mythos 5 (the unclassified variant) was available only to a small group of pre-approved Claude Mythos Preview customers and is also currently suspended.
• Regional availability: At launch, Fable 5 supported In-Region, Geographic (Geo), and Global inference routing on Amazon Bedrock; specific region list is detailed in the Amazon Bedrock regional availability documentation.
• GovCloud: Access to Bedrock foundation models in AWS GovCloud requires EULA acceptance in a standard region (us-east-1 or us-west-2) before enabling in the GovCloud account.
• Pricing: Consumption-based pricing per token; prompt-stage refusals are not billed; mid-stream refusals are billed for tokens generated before the block; refer to the Amazon Bedrock Pricing page for current rates.
• Data retention requirement: A mandatory 30-day data retention policy applies to Fable 5 on Amazon Bedrock; Zero Data Retention (ZDR) is available on Claude Platform on AWS by request through an Anthropic account representative.
• Knowledge cutoff: January 2026.

Related Resources

• https://www.anthropic.com/news/fable-mythos-access
• https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-anthropic-claude-fable-5.html
• https://docs.aws.amazon.com/bedrock/latest/userguide/models-region-compatibility.html#model-regions-anthropic
• https://docs.aws.amazon.com/claude-platform/latest/userguide/welcome.html