Live Nov 18 · Cohort 3Nov 18, 19:00 MSK — free hands-on AI agents live workshop · Live Cohort 3 starts Nov 23Join Nov 18 Live →Cohort 3 (Nov 23)
ReleaseAPI / Product · ActionableMarkTechPost→ original

Anthropic releases Claude Fable 5 and Mythos 5: one model, two access tiers

Anthropic has released two variants of the same base model. Claude Fable 5 is publicly available with the full stack of safety classifiers. Claude Mythos 5 is the same model with its cybersecurity restrictions deliberately removed, and only members of the closed Project Glasswing program can use it. With this release, Anthropic formalizes a new Mythos-class tier for trusted security partners.

AI-processed from MarkTechPost; edited by Hamidun News
Anthropic releases Claude Fable 5 and Mythos 5: one model, two access tiers
Source: MarkTechPost. Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourceMarkTechPost ↗
Editor-in-Chief
Key metric #1Offensive security researchers testing real-world attack vectors
Key metric #2Red team specialists working on system vulnerabilities
Key metric #3Government cyber defense agencies

Anthropic has made Claude Fable 5 generally available. Alongside it, the company announced Claude Mythos 5, a version of the same base model with a fundamentally different safety profile. Mythos 5 is available exclusively through Project Glasswing, a closed program.

Two products, one core

The key detail of this release is that Fable 5 and Mythos 5 share the same base weights. They differ neither in architecture nor in training. The only difference is the set of classifiers and safety policies layered on top.

Claude Fable 5 ships to the public in its standard configuration, with a full stack of protective filters tuned for a broad audience. This is the version for API developers, enterprise customers, and end users, and it goes through Anthropic's standard safety evaluation process.

Claude Mythos 5 is a different story. It is the same model, but with its cybersecurity restrictions deliberately removed. It will not be sold on the open market.

Project Glasswing: a closed channel

Project Glasswing is Anthropic's program for trusted security partners. Through it, the company gives selected organizations access to a model with expanded capabilities in areas where standard restrictions would get in the way of professional work.

Typical program participants include:

  • Offensive security researchers testing real-world attack vectors
  • Red team specialists working on system vulnerabilities
  • Government cyber defense agencies
  • Corporate security teams at major infrastructure companies
  • Academic groups studying AI threats

Access to the program is tightly controlled: Anthropic vets partners before granting them Mythos 5.

A new tier in the model hierarchy

With Mythos 5, Anthropic formalizes a new level in its product lineup: Mythos-class. This isn't a marketing label. It's a separate deployment mode with different access terms, dedicated documentation, and usage monitoring on Anthropic's side.

The hierarchy used to be simple: Haiku, Sonnet, Opus. Fable 5 has taken the spot of top public model. Mythos-class adds an entirely new dimension, defined not by capability but by usage policy.

"Same model, different safeguards" is a statement of approach: safety as a configuration

layered on top of the base weights, not as an intrinsic property of the model itself.

A response to pressure from the security market

Anthropic has long faced criticism that its models are too cautious in exactly the places where security professionals want to work with real threats. Competing options, whether open models or less strict providers, capture this segment simply because they impose fewer restrictions.

Project Glasswing and Mythos 5 are a direct response to that challenge. Anthropic isn't lowering the bar for all users. Instead, it is building a controlled channel for professionals who need capabilities beyond the standard profiles.

What it means

Anthropic is betting on segmentation: the broad market gets Fable 5 with full protections, while trusted security partners get Mythos 5 with a free hand. It's an attempt to avoid ceding the security market to competitors while preserving the company's reputation for prioritizing safe AI. The open question is how strict control over Glasswing participants will remain as the program grows.

Story

Key Questions Answered

What are the key details on Two products, one core?

The key detail of this release is that Fable 5 and Mythos 5 share the same base weights. They differ neither in architecture nor in training. The only difference is the set of classifiers and safety policies layered on top. Claude Fable 5 ships to the public in its standard configuration, with a full stack of protective filters tuned for a broad audience. This is the version for API developers, enterprise customers, and end users, and it goes through Anthropic's standard safety evaluation process.

What are the key details on Project Glasswing: a closed channel?

Project Glasswing is Anthropic's program for trusted security partners. Through it, the company gives selected organizations access to a model with expanded capabilities in areas where standard restrictions would get in the way of professional work.

What are the key details on A new tier in the model hierarchy?

With Mythos 5, Anthropic formalizes a new level in its product lineup: Mythos-class. This isn't a marketing label. It's a separate deployment mode with different access terms, dedicated documentation, and usage monitoring on Anthropic's side. The hierarchy used to be simple: Haiku, Sonnet, Opus. Fable 5 has taken the spot of top public model. Mythos-class adds an entirely new dimension, defined not by capability but by usage policy.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Unified AI Workspace · Nomad

Test Frontier LLMs & Autonomous Agents in One Workspace

Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.

Try Nomad AI →Launch Nomad Agent
What do you think?
Loading comments…