Anthropic Released the AI Model It Was Too Scared to Give Everyone
Claude Fable 5 gives the public Anthropic’s newest frontier model, while Mythos 5 stays locked behind trusted access for cyber and infrastructure users.
Anthropic has officially released Claude Fable 5, its most powerful AI model available to the public. But the real story is not just that the model is smarter. It is that Anthropic is now openly splitting frontier AI into two worlds: the version everyone can use, and the version only trusted organizations are allowed to touch.
Claude Fable 5 is part of Anthropic’s new Mythos-class family, a tier that sits above its Opus models. According to Anthropic, Fable 5 and Claude Mythos 5 can work autonomously for longer than any previous Claude models, with stronger performance across software engineering, knowledge work, vision, memory, and life sciences research.
That sounds like a normal AI launch until you get to the weird part.
Fable 5 is not the full unlocked model.
Anthropic is also releasing Claude Mythos 5, which is built on the same underlying model as Fable 5 but with some safety restrictions lifted. Mythos 5 is not generally available. It is being limited to cyberdefenders, infrastructure providers, and selected organizations through Anthropic’s Project Glasswing program.
In other words, Anthropic has released one model for the public and another version for people it trusts.
Why Fable 5 Matters
Anthropic says Fable 5 is made for long-horizon work. That means it is not just answering questions or writing quick code snippets. It is supposed to handle bigger tasks that require planning, memory, reasoning, tool use, and sustained focus over a long period.
The technical numbers make that clearer. According to Anthropic’s Claude API docs, Claude Fable 5 and Claude Mythos 5 support a 1 million token context window on the Claude API by default, and a single request can generate up to 128,000 output tokens. Anthropic’s API release notes also say both models support always-on adaptive thinking.
That makes Fable 5 useful for huge code migrations, massive document analysis, research workflows, complex debugging, and multi-step agent tasks that cheaper models might struggle to hold together.
For developers, the most interesting part is probably coding. Anthropic says the model is especially strong at software engineering and long-running agentic work. The company is clearly positioning it as more than a chatbot. This is an agent model.
That is also why the safety discussion matters.
A model that can understand massive codebases, reason across long tasks, and operate semi-autonomously is useful for building software. But the same capabilities can also be useful for finding vulnerabilities, writing exploits, or helping attackers move faster.
That is the central tension of Fable 5.
The Safeguards Are the Product
Anthropic’s pitch is simple: Fable 5 brings Mythos-level intelligence to general users, but with guardrails.
When Fable 5 detects certain high-risk requests, especially around cybersecurity, biology, chemistry, or model distillation, it can refuse the request or route the user to Claude Opus 4.8 instead. Anthropic says more than 95 percent of Fable 5 sessions do not fall back to Opus 4.8, but the company also admits the safeguards may sometimes block harmless requests.
That is a huge detail.
The model is not just being judged by how smart it is. It is being judged by whether Anthropic can control where that intelligence gets applied.
This is the new frontier AI problem. The race is no longer only about who has the smartest model. It is about who can release powerful models without turning them into scalable weapons.
That is also the angle The Verge focused on: Anthropic is releasing a model class it previously considered too risky for public access, now wrapped in new safeguards.
Mythos 5 Is the More Sensitive Release
Claude Mythos 5 is the version Anthropic is not giving everyone.
It has the same underlying model as Fable 5, but with some safeguards lifted for approved users. Anthropic says Mythos 5 will initially be deployed through Project Glasswing, its cybersecurity-focused initiative involving major tech, cloud, infrastructure, and security organizations.
That program exists because Anthropic’s earlier Mythos Preview model showed serious cybersecurity capabilities. Anthropic has said Mythos-class models can help defenders find vulnerabilities and secure critical systems, but the obvious fear is that the same ability could help attackers too.
This is why Anthropic is treating Mythos-class models differently. These models are not just better at writing essays or generating code. They appear to cross a threshold where they can meaningfully help with vulnerability discovery, security research, and potentially offensive cyber work.
For defenders, that is powerful.
For attackers, it is terrifying.
The Price Also Says a Lot
Fable 5 is expensive.
Anthropic is charging $10 per million input tokens and $50 per million output tokens. That is double the standard pricing of Claude Opus 4.8, which sits at $5 input and $25 output per million tokens.
That price makes the positioning obvious. Fable 5 is not meant to replace cheaper models for everyday tasks. You would not use it for simple summaries, basic support bots, or routine content generation unless cost does not matter.
It is meant for the hard stuff: long coding tasks, enterprise analysis, scientific research, finance workflows, vision-heavy reasoning, and agent systems where one good answer is worth the extra cost.
For most companies, Sonnet or Opus will still make more sense for normal workloads. Fable is the expensive model you call when the task is messy, long, high-value, or too complex for cheaper models to handle reliably.
The Bigger Picture
The timing is important too.
Anthropic is releasing Fable 5 while the AI industry is shifting from “chatbots that answer questions” to “agents that do work.” The company is also expanding Claude deeper into enterprise workflows, where models are expected to inspect codebases, manage documents, use tools, and complete longer tasks with less supervision.
Fable 5 fits perfectly into that story. It is not just another model release. It is Anthropic showing what the next generation of AI products may look like: more autonomous, more capable, more expensive, and more tightly controlled.
The scary part is that Anthropic is basically saying the quiet part out loud. The best AI models are becoming powerful enough that simply releasing them openly may no longer be responsible.
So Fable 5 is both a launch and a compromise.
It gives the public access to Anthropic’s most capable widely released model yet, but only after safety systems are placed between the user and the model’s full power. Meanwhile, the less restricted version stays behind a trust gate.
That may become the new normal for frontier AI.
Not one model for everyone.
One model for the public.
Another model for the trusted few.
