Overview
Claude Mythos Preview is a general-purpose frontier model from Anthropic, the first member of its Claude Fable / Mythos (Mythos-class) line. Anthropic announced it on April 7, 2026, alongside Project Glasswing. Unlike Anthropic's widely released Opus and Sonnet models, Mythos Preview was never made generally available — Anthropic explicitly stated it did not plan to release it to the public, and distributed it only to vetted partners.
Mythos Preview is a broadly strong model that proved strikingly capable at computer-security tasks. Working through Project Glasswing, Anthropic and its partners used Claude Mythos Preview to find more than ten thousand high- or critical-severity vulnerabilities across systemically important software, including zero-day flaws in every major operating system and every major web browser. Anthropic notes these cybersecurity abilities emerged as a byproduct of improvements to coding and reasoning rather than from training the model specifically for vulnerability discovery.
Claude Mythos Preview was succeeded on June 9, 2026 by Claude Mythos 5 (the same restricted Project Glasswing distribution) and Claude Fable 5 (the publicly available variant with extra safety classifiers). Anthropic says users find Mythos 5 comparable to — or somewhat stronger than — Mythos Preview in most cases, while costing substantially less.
| Released | 2026-04 |
|---|---|
| License | Proprietary |
| Weights | API only |
| Modalities | Text |
| Status | Superseded |
Pricing
| Input | $25 / 1M tokens per 1M tokens |
|---|---|
| Output | $125 / 1M tokens per 1M tokens |
Pricing quoted by Anthropic for access following the research-preview period, via the Claude API, Amazon Bedrock, Google Vertex AI, and Microsoft Foundry. Its successors Fable 5 and Mythos 5 are priced lower, at $10/$50 per 1M tokens — described by Anthropic as less than half the price of Mythos Preview.
Strengths
- Autonomous discovery of zero-day software vulnerabilities across major operating systems and browsers
- Building working exploits by chaining multiple small attack primitives together
- Reverse engineering closed-source binaries and turning known (N-day) flaws into functional exploits
- Strong general-purpose coding and reasoning, from which its security capabilities emerged
- Long-horizon, autonomous agentic security work over large codebases
Best for
- Defensive vulnerability scanning of critical software and infrastructure codebases
- Security research and exploit-chain analysis by vetted partners
- Reverse engineering and binary analysis
- Hardening systemically important open-source and commercial software
- Cyberdefense programs run with infrastructure providers and government partners
How to access
| Provider | Model ID |
|---|---|
| Anthropic (Claude API) ↗ | claude-mythos-preview |
| Amazon Bedrock ↗ | — |
| Google Vertex AI ↗ | — |
| Microsoft Foundry ↗ | — |
Claude Fable / Mythos (Mythos-class) — every version
The full lineage of the Claude Fable / Mythos (Mythos-class) line, newest first. Every version has its own page — click any to compare specs, benchmarks and pricing.
| Version | Released | Context | License |
|---|---|---|---|
| Claude Fable 5current | 2026-06-09 | — | Proprietary |
| Claude Mythos 5 | 2026-06-09 | — | Proprietary |
| Claude Mythos Preview | 2026-04 | — | Proprietary |
FAQ
Is Claude Mythos Preview available to the public?
No. Anthropic stated it did not plan to make Claude Mythos Preview generally available. Access was invitation-only through Project Glasswing — initially around 40-50 partners (including Microsoft, Apple, Google, AWS, the Linux Foundation, Cisco, and Nvidia), later expanded to roughly 150-200 organizations across more than 15 countries.
What is Claude Mythos Preview used for?
It is a general-purpose model that proved exceptionally capable at cybersecurity. Project Glasswing partners used it defensively to scan critical codebases, where it found more than ten thousand high- or critical-severity vulnerabilities, including zero-days in every major operating system and web browser.
How does Mythos Preview relate to Claude Fable 5 and Mythos 5?
Mythos Preview is the first model in Anthropic's Mythos-class line. It was succeeded on June 9, 2026 by Mythos 5 (still restricted to Project Glasswing) and Fable 5 (the public variant with additional safety classifiers). Anthropic says Mythos 5 is comparable to or somewhat stronger than Mythos Preview, at substantially lower cost.
How much did Claude Mythos Preview cost?
Anthropic quoted $25 per million input tokens and $125 per million output tokens for access following the research-preview period, available via the Claude API, Amazon Bedrock, Google Vertex AI, and Microsoft Foundry. The later Fable 5 and Mythos 5 models are priced lower at $10/$50 per million tokens.