Guide

Claude Mythos explained: what it is and whether you can use it

Published 30 September 2026

A heavy steel vault door standing slightly open with soft blue light spilling out, in a quiet modern server room

Short answerClaude Mythos is Anthropic's most capable model for cybersecurity and biology research, and it is not open to the public. Approved organisations get it through Project Glasswing. Everyone else can use its twin, Claude Fable 5.1, which has the same capabilities plus safety filters.

If you have seen "Claude Mythos" in a headline and wondered whether you can try it, here is the short answer: almost certainly not, and you probably don't need to. This guide explains what Mythos is, why Anthropic keeps it locked down, what it costs, and which model you can use right now to get the same abilities.

What Claude Mythos is

Anthropic describes Claude Mythos as its "most capable model for cybersecurity and biology research". The latest version is Claude Mythos 5.1, announced on 1 September 2026. It is built for work such as finding software vulnerabilities, threat intelligence, red teaming, drug discovery and biodefence screening.

The name has been around since spring. According to Wikipedia's summary of the timeline, Anthropic first disclosed Claude Mythos Preview on 7 April 2026, followed by Mythos 5 on 9 June and Mythos 5.1 on 1 September. The Preview was not released to the public, and neither were the versions after it.

Anthropic keeps Mythos restricted because "it could be used both for good and for harm." A model that is very good at finding weaknesses in software can help defenders fix them, and it can also help attackers.

So access works through approved programmes instead of a sign-up page:

Project Glasswing is Anthropic's programme for using Mythos to scan critical software for vulnerabilities. Anthropic says it was extended on 2 June 2026 to about 150 new organisations in more than 15 countries. Wikipedia lists partners including Microsoft, Apple, Google, Amazon Web Services, the Linux Foundation, Cisco and Nvidia.
The Cyber Verification Program is for defensive security professionals, and the Life Sciences Verification Program is for advanced life-science researchers.
Anthropic says it can currently offer Mythos 5.1 only to a set of US organisations, though it is working to expand that.

If you work at a company that already has an Anthropic, AWS or Google Cloud account team, Anthropic's documentation says to contact them for access. If you are a student, a freelancer or a small business, this is not a route that is open to you today.

What it costs

For the organisations that do get access, Anthropic lists the same price for Mythos 5.1 as for Fable 5.1:

	Claude Mythos 5.1	Claude Fable 5.1
Who can use it	Project Glasswing participants only	All customers
API name	`claude-mythos-5-1`	`claude-fable-5-1`
Input price	$10 per million tokens	$10 per million tokens
Output price	$50 per million tokens	$50 per million tokens
Context window	1 million tokens	1 million tokens
Maximum reply	128,000 tokens	128,000 tokens

Both models keep your data for 30 days by default for safety monitoring, and they aren't offered under zero data retention unless Anthropic expressly agrees.

Mythos vs Fable: what's the difference?

This is the part most headlines skip. Anthropic's documentation says Claude Fable 5.1 has the same capabilities as Mythos 5.1, and that Fable is the version "with safeguards for broader availability."

In practice, that means Fable has safety classifiers that can decline a request. When it does, the API returns a normal response with stop_reason: "refusal" and says which category was triggered (for example cyber or bio). Anthropic's Mythos 5 documentation says Mythos does not include those classifiers, which is why it is limited to vetted organisations.

For everyday work, such as long research, big coding projects, spreadsheets and slide decks, Fable 5.1 does the same job and is open to everyone. Anthropic calls it its "most capable model open to all customers."

What to use instead

Here is a simple way to choose, based on Anthropic's own model-choosing guide:

You want the strongest model that anyone can use: Claude Fable 5.1. It is built for long, demanding work, such as agent sessions that run for hours and multi-step research.
You want a strong all-rounder that costs less: Claude Opus 5.5 at $4 per million input tokens. Anthropic says most workloads should start here.
You want speed and low cost for everyday tasks: Claude Sonnet 5.5 at $2 per million input tokens. See our guide to using Sonnet 5.5.

Our comparison of every current Claude model puts them side by side.

Three things to be careful about

"Get Mythos access" offers. Any site that sells Mythos access or asks for your Claude login is not from Anthropic. Real access goes through the verification programmes.
Big numbers. You will see parameter counts and vulnerability totals quoted in articles. Some come from Anthropic, some from secondary sources. Treat any number without a link to Anthropic as unconfirmed.
Security work needs permission. Even if you did get access, only test systems you own or are authorised to test.

Ready-made prompts for reading about new AI models

New models arrive every few weeks, and it helps to have a repeatable way to make sense of the announcements. These PromptUp cards work in any assistant:

Paste an announcement into the executive summary prompt to get a headline, key points and next steps.
The explain it simply prompt turns technical release notes into plain words.
The competitor comparison prompt sets two or three models side by side.

A copy-ready prompt for any model announcement:

prompt — 1 blank

I'm going to paste an AI company's announcement for a new model.

Task: Tell me in plain words what it is, who can use it today, and what it costs.
Done means: 5 bullets, each under 25 words, then one line saying whether an ordinary user needs to care.
Rules: Use only what's in the text below. If access or price isn't stated, write "not stated" instead of guessing. Separate what the company claims from what has been independently checked.

[PASTE THE ANNOUNCEMENT HERE]

Where to go from here

If you want to get the most from the Claude models you can use, read our guides to prompting Claude Opus 5.5, using Claude Sonnet 5.5 and prompting Claude Fable 5.1. The Claude prompts collection has hundreds of free, tested prompts, and the prompt builder can fill in a template for you.

Quick checklist

Mythos is for approved cybersecurity and life-science organisations, not the public.
Access goes through Project Glasswing and Anthropic's verification programmes.
Claude Fable 5.1 has the same capabilities, with safety filters, and anyone can use it.
Most people should start with Sonnet 5.5 or Opus 5.5 and only move up if the work needs it.
Be wary of any site that claims to sell Mythos access.

Model facts change quickly. Check Anthropic's model documentation before you make a purchase or a plan based on anything here.