Anthropic restricted model · checked July 30, 2026 Invitation only · trusted access

Quick answer

Mythos 5 is a trusted-access profile, not an upgrade button.

It shares Fable 5's underlying model and specifications but lifts safeguards for approved defensive programs. The model remains invitation-only with no self-serve sign-up.

Verified provider record

Provider status

Limited availability for approved Project Glasswing and trusted-access organizations; there is no self-serve sign-up.

Claude API ID

claude-mythos-5. The identifier does not grant access without organization approval and provisioning.

Underlying model

The same underlying model, specifications, and published pricing as Claude Fable 5, with safeguards lifted in approved domains.

Context and output

Anthropic says Mythos shares Fable's specifications: 1 million token context and up to 128,000 output tokens.

Published API price

$10 per million input tokens and $50 per million output tokens for approved access. Verify the governing contract.

Flowith access

Not asserted here. Restricted provider access does not establish a Flowith model mapping.

Access boundary

Approval is part of the model contract

General users

Choose Fable 5 or another generally available Claude model. Mythos is not an upgrade selectable by a consumer plan.

Cyber defenders

Mythos is deployed through Project Glasswing and approved trusted access for defensive cybersecurity organizations.

Biology researchers

Anthropic described a separate small trusted program using Fable with biology and chemistry safeguards removed while cyber safeguards remain.

Cloud customers

Contact the Anthropic, AWS, or Google Cloud account team; a visible model name or ID is not proof of entitlement.

Governance checks before access

  • Mythos removes safeguards only within the approved program and domain; it is not an unrestricted public model.
  • Access terms, reviewer scope, acceptable use, logging, incident response, and human oversight belong in the organization agreement.
  • Covered models require 30-day prompt and output retention for organizations that otherwise use zero-data-retention arrangements.
  • Anthropic says retained data is for safety work, controlled human review is logged, and deletion occurs after 30 days in almost all cases.
  • Do not infer that Mythos access, cloud provisioning, or provider pricing makes the model available in Flowith.

Official sources

  1. Anthropic launch — shared model, safeguards, trusted access, pricing, and initial availability.
  2. Anthropic redeployment update — current restricted restoration and safeguard revisions.
  3. Claude model overview — invitation-only status, IDs, shared specifications, and account-team route.
  4. Covered-model retention policy — scope, 30-day retention, review controls, and exceptions.

Claude Mythos 5 questions, answered

Claude Mythos 5 is Anthropic's restricted profile of the same underlying model used by Claude Fable 5. It has safeguards lifted in approved domains and is offered only through Project Glasswing and trusted-access programs.
No. Anthropic's current model documentation says Mythos 5 is invitation-only with no self-serve sign-up. Approved organizations must work through Anthropic or a supported cloud account team.
Anthropic names claude-mythos-5, but the identifier is usable only after the organization receives approval, provisioning, and the relevant platform entitlement.
Anthropic says they share the same underlying model, specifications, and published pricing. The primary difference is the safeguard and access profile.
For organizations using zero-data-retention arrangements, Anthropic's covered-model policy requires prompts and outputs to be retained for 30 days for safety work. Consumer retention is governed separately and was already in place.
This page does not claim a Flowith mapping. Mythos provider access is restricted, and any Flowith availability would require separate first-party confirmation.