Breaking

Anthropic Launches Claude Fable 5, Its Top-Tier Model, for General Use

June 9, 2026 at 14:28 EDT

Foundation Models
AI Agents
Software Dev & Coding

On June 9, 2026, Anthropic announced Claude Fable 5 (model ID: claude-fable-5), the first publicly available Mythos-class model, rolling it out the same day across Claude API, AWS, Amazon Bedrock, Vertex AI and Microsoft Foundry. A higher-tier claude-mythos-5 is being offered in limited fashion through Project Glasswing.

June 9, 2026 · Anthropic

Claude Fable 5 Launches — the "Safe" Twin of a More Dangerous Mythos

The first public model in Anthropic's new Mythos-class generation arrives on Web, CLI and Desktop. Its more capable sibling, Mythos 5 — armed with advanced cyber abilities — stays locked behind the trusted "Glasswing" program.

<5%

Average rate that queries trigger the high-risk safety route

Surfaces at launch — Web, CLI & Desktop app

Shared base model behind both Fable 5 and Mythos 5

Two roles, one base model

Same engine — opposite levels of restraint.

Claude Fable 5

Public — available on all plans
Strengthened safety guardrails
Civilian / general tasks
The "leashed" safe version

Claude Mythos 5

Glasswing partners only
Relaxed restrictions
Advanced cyber (zero-day discovery)
The more dangerous sibling

How the safeguard works

High-risk queries are automatically re-routed to Claude Opus 4.8.

Incoming query

→

Risk check
cyber · bio · chem · distillation

→

Fallback to
Opus 4.8

Triggers on average less than 5% of the time — but harmless queries can sometimes be falsely flagged.

Praise

"Most powerful public model right now"
Strong on long-running tasks & coding agents
Excels at long-document & image analysis, spreadsheets

Complaints

Frequent access errors at launch
Safety filter over-refuses routine bug fixes
Tight context/usage limits, weak instruction following
The "nerfed" twin — real power lives in Mythos 5

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.