← Models

Claude Mythos 5

Anthropic · Claude Mythos

Anthropic's limited-availability Mythos model for approved Project Glasswing defensive cybersecurity workflows.

Type
multimodal
Context
1M tokens
Max Output
128K tokens
Status
preview
Input
$10/1M tok
Output
$50/1M tok
API Access
Yes
License
proprietary
preview cybersecurity coding agentic long-context defensive-security
Released June 2026 · Updated July 6, 2026

Overview

Freshness note: Model capabilities, limits, pricing, and availability can change quickly. This profile is a point-in-time snapshot last verified on July 6, 2026.

Claude Mythos 5 is Anthropic’s limited-availability Mythos route for approved Project Glasswing customers. It was released on June 9, 2026 alongside Claude Fable 5, then restored for selected US organizations on July 1 after Anthropic updated Fable safeguards and coordinated with government and industry partners.

This is not a general-purpose self-serve model page in practice. Anthropic’s current model docs say Mythos 5 shares Fable 5 specs and pricing, but that it is not generally available. It is offered only to approved customers for defensive cybersecurity work through Project Glasswing.

Capabilities

Anthropic frames Mythos 5 as the restricted version of the Fable/Mythos capability tier. The public materials say Fable 5 and Mythos 5 share the same underlying model, but Mythos has fewer safeguards and is intended for defensive cybersecurity partners that need higher latitude for vulnerability discovery, security analysis, and related protected work.

That makes the model strategically important even if most teams cannot use it. It signals the frontier of Anthropic’s cyber-capable agent work, while Fable 5 and Opus 4.8 remain the more realistic routes for ordinary enterprise and software teams.

Technical Details

Current official anchors:

  • API ID: claude-mythos-5
  • 1M-token context window
  • 128K max output
  • text and image input, text output
  • adaptive thinking support
  • limited availability for approved Project Glasswing customers

Anthropic’s model overview also keeps claude-mythos-preview as a separate invitation-only model. Treat Mythos 5 and Mythos Preview as related but distinct access routes, and verify the actual model ID in any approved deployment.

Pricing & Access

Published Mythos 5 API pricing matches Fable 5:

  • Input: $10 per 1M tokens
  • Output: $50 per 1M tokens
  • 5-minute cache writes: $12.50 per 1M tokens
  • 1-hour cache writes: $20 per 1M tokens
  • Cache hits and refreshes: $1 per 1M tokens
  • Batch API input: $5 per 1M tokens
  • Batch API output: $25 per 1M tokens

Access is the limiting factor. Anthropic says Mythos 5 is invitation-only, has no self-serve sign-up, and requires coordination through Anthropic, AWS, or Google Cloud account teams for approved Project Glasswing use.

Best Use Cases

Mythos 5 belongs in tightly governed defensive cybersecurity programs: vulnerability discovery, patch validation, secure-code review, security research, and critical-system hardening where the organization can justify the access controls and review burden.

It is not appropriate as a default assistant, public product backend, or casual coding model. Teams outside the approved program should route to Sonnet 5, Opus 4.8, or Fable 5 depending on cost, safeguards, and capability needs.

Comparisons

  • Claude Fable 5 (Anthropic): Same published specs and pricing, but Fable is the broadly released route with stronger safeguards.
  • Claude Mythos Preview (Anthropic): Earlier invitation-only Mythos surface tied to Project Glasswing; Mythos 5 is the newer named model route.
  • Claude Opus 4.8 (Anthropic): More broadly available premium route for hard coding and agentic work.
  • GPT-5.5 (OpenAI): Broader frontier generalist, while Mythos 5 is narrower and governed around defensive cybersecurity access.