Category: AI Tools | Anthropic | Claude | Tech News
Published: June 14, 2026
Read time: 7 min
Two months ago, Anthropic released a model so powerful that they would only give it to a handful of vetted organizations. It found zero-day vulnerabilities in every major operating system and browser. It discovered a 27-year-old security flaw in OpenBSD. It autonomously wrote a working exploit against a 17-year-old FreeBSD bug — giving an unauthenticated attacker full root access from anywhere on the internet.
That model was called Mythos. And on June 9, 2026, Anthropic made a version of it available to everyone.
Claude Fable 5 is Anthropic’s most capable AI model ever released to the public — a Mythos-class model with safety guardrails added, now available through the Claude API, Claude.ai, Claude Code, and enterprise plans. Here is everything you need to know.
What Is Claude Fable 5?
Fable 5 is best understood as Mythos’s public twin. Both models share the same underlying capability — the most powerful AI Anthropic has ever built. The difference is a layer of safety classifiers that distinguishes what each model will and will not do.
Anthropic even explains the naming: Fable comes from the Latin fabula — “that which is told” — which is akin to the Greek mythos. Same idea, different safety profile.
Here is how the two models compare:
| Claude Fable 5 | Claude Mythos 5 | |
|---|---|---|
| Availability | Public — API, Claude.ai, Enterprise | Restricted — Project Glasswing vetted orgs only |
| Cybersecurity capabilities | Blocked — falls back to Opus 4.8 | Full access for vetted users |
| Biology/Chemistry | Blocked for high-risk tasks | Available for approved organizations |
| Pricing | $10/M input, $50/M output tokens | $10/M input, $50/M output tokens |
| Context window | 1 million tokens | 1 million tokens |
| Max output tokens | 128,000 | 128,000 |
| Knowledge cutoff | January 2026 | January 2026 |
Both models are priced at less than half the cost of the earlier Mythos Preview — making Fable 5 not just more accessible, but significantly more affordable for developers and businesses.
What Makes It So Capable?
Anthropic describes Fable 5 as built for “days-long, complex, and asynchronous tasks that previous models could not sustain.” That is a meaningful claim — and it reflects several genuine capability advances.
Extended agentic execution. Fable 5 can run multi-step autonomous workflows over extended periods with minimal human intervention. Tasks that previously required constant oversight — like migrating a 50-million-line codebase or building production-ready applications — are now within its sustained capability range.
Advanced vision understanding. The model handles complex visual inputs with significantly improved accuracy, including diagrams, nested tables, charts, and technical documents that require understanding spatial relationships, not just text.
Self-verification behaviors. Fable 5 can check its own work during execution — catching errors, revising approaches, and flagging uncertainties rather than confidently producing incorrect outputs.
Superior software engineering. Anthropic specifically highlights software engineering, knowledge work, and vision as Fable 5’s strongest areas. For developers using Claude Code, this translates directly into more capable, more reliable agentic coding assistance.
The Safety Architecture — How the Guardrails Work
This is where Fable 5 gets genuinely interesting from a technical standpoint.
When a prompt triggers Fable 5’s safety classifiers — for cybersecurity, biology, chemistry, or distillation topics — the model does not simply refuse. Instead, it falls back to Claude Opus 4.8 for that specific request. The user gets a response, just from a less capable model for that particular task.
Technically, when a refusal occurs, the API returns stop_reason: "refusal" as an HTTP 200 response — and critically, the user is not billed if no output is generated.
This architecture is deliberate. Anthropic is not trying to make Fable 5 completely incapable in sensitive areas — it is trying to ensure that the most dangerous applications remain available only to vetted organizations through Mythos 5, while legitimate users can still get help with general work in adjacent fields.
However, as cybersecurity researchers have noted, the current classifiers can be overly broad — flagging requests like “write secure code” or “do a code review” as cybersecurity-related, when they are really just standard software engineering. Anthropic has acknowledged this and indicated the guardrails will be refined over time. Professionals who need fuller access can apply to Anthropic’s Cyber Verification Program for expanded capabilities.
Pricing and Access — What You Need to Know
Current availability:
- Claude API — available now via
claude-fable-5 - Claude.ai — available on Pro, Max, Team, and Enterprise plans
- Claude Code — available via web and CLI
- AWS Bedrock — available
- GitHub Copilot — available (requires up to 30 days of prompt retention for safety classifiers)
Pricing:
- $10 per million input tokens
- $50 per million output tokens
- 90% discount on input tokens with prompt caching
- US-only inference available at 1.1x pricing
Important pricing note: Fable 5 is currently included in subscription plans until June 22, 2026. After that date, the cost doubles compared to Opus — reflecting its significantly advanced capabilities. If you want to explore the model before the pricing change, now is the time.
Fable 5 vs Mythos 5 — Which One Is for You?
The vast majority of users and businesses will find Fable 5 fully sufficient for their needs. The capability difference between Fable 5 and Mythos 5 is not about general intelligence — it is specifically about access to advanced offensive cybersecurity capabilities.
Choose Fable 5 if you are:
- A developer building applications, automating workflows, or working with Claude Code
- A business using AI for content, analysis, research, or knowledge work
- An enterprise deploying AI across large-scale, long-running tasks
- Anyone using Claude for general professional or creative work
Mythos 5 is for:
- Vetted cybersecurity organizations through Project Glasswing
- Critical infrastructure operators approved by Anthropic
- Organizations specifically defending against advanced cyber threats
For TheTechCursor readers — marketers, SEO professionals, digital marketing teams, and developers — Fable 5 is the most capable AI tool ever made publicly available, and it is accessible right now.
The Bigger Context — Why This Launch Matters
Fable 5’s release comes at a remarkable moment. Anthropic has simultaneously:
- Confidentially filed its S-1 with the SEC, preparing for a public offering at a ~$965 billion valuation
- Warned publicly that AI systems are advancing so rapidly they may soon achieve recursive self-improvement — autonomously improving themselves without human intervention
- Released its most powerful model yet to the general public
That combination — a public safety warning alongside a public model release — reflects the central tension Anthropic navigates constantly. The company believes frontier AI is potentially dangerous. Nevertheless, it also believes that safety-focused organizations releasing capable models is better than ceding the frontier to less safety-conscious competitors.
Fable 5 is the clearest expression of that philosophy yet: maximum capability, with the most dangerous applications carefully gated rather than entirely withheld.
Bottom Line
Claude Fable 5 is the most capable AI model Anthropic has ever made publicly available. For developers, it brings sustained agentic execution, advanced vision, and self-verification to complex, long-running tasks. For businesses, it represents a step-change in what AI can autonomously handle at scale.
The pricing window before June 22 makes right now the best time to explore what Fable 5 can do for your work. After that, it will cost significantly more — but given what it can do, for many use cases, it will be worth it.