⚫ In memoriam — Joshua Baer, founder of Capital Factory (1975–2026). Read the tribute →
ai·wholetech·com
ai.wholetech.com / Assistants / Claude / Mythos 5
⌁ Getting-started guide · updated June 10, 2026

Claude Mythos 5 — what it is, and how to actually start.

Anthropic's new top model comes in two forms: Mythos 5, the unrestricted version for vetted scientists and cyber defenders, and Fable 5, the same brain made safe for everyone. This guide covers both — what Mythos is, who gets it, and the step-by-step path to using the Mythos-class model that's available to you today.

One page, plain English, no jargon. Manual · resources · videos · courses · Austin training.

Same model, two doors Mythos: invitation only Fable: available now $10 / $50 per M tokens
If you remember one thing

Mythos 5 is invitation-only. The way to use it today is Fable 5 — the exact same model with safety guardrails on, available to everyone. Open your model menu, pick Fable 5, and you're running Mythos-class AI.

01Two doors into one model

On June 9, 2026, Anthropic shipped a new model family it calls "Mythos-class" — a tier above its previous flagship, Opus. Rather than choosing between shipping it widely or keeping it safe, Anthropic did both, through two doors:

Fable 5 — the public door

Everyone · available now

The full Mythos-class model with safety classifiers in front of it. A narrow set of risky topics (advanced cybersecurity, biology/chemistry, attempts to copy the model) get quietly answered by Opus 4.8 instead. Anthropic says this happens in under 5% of sessions — for normal work you'll never notice. Everything in section 03 below is about getting started with Fable.

Mythos 5 — the vetted door

Restricted · partners & researchers

The same model with those guardrails lifted. Today it's limited to Project Glasswing participants — cyber defenders and infrastructure providers. A biology-researcher program (cyber safeguards kept, bio protections lifted) launches in the coming weeks, and a formal application process is planned but not yet published.

Why bother locking up Mythos? Because the capability is genuinely dual-use. The early results Anthropic reported — roughly 10× faster drug design (9 of 14 protein targets produced strong candidates), molecular-biology hypotheses scientists preferred 80% of the time in blinded tests, and an unattended week of genomics work that beat a recent Science paper — come from the same machinery that could, in the wrong hands, help design something harmful. The condensed announcement breakdown lives on our Fable 5 page.

02Can you get Mythos 5 itself?

Honest answer for almost everyone: not yet, and possibly not ever directly — and that's fine, because Fable 5 is the same model for everyday work. Here's the current access picture:

Who should pursue it: if you run a research lab, a biotech, or critical infrastructure security, the coming application program is worth tracking. If you're building websites, writing, analyzing documents, or coding — which is most of us — Fable 5 is your Mythos. Skip ahead.

03Getting started today — the manual

Fable 5 is rolling out across subscription plans June 9–23, so if you don't see it in a menu yet, check back in a day or two — that's expected, not a problem. Three ways in, easiest first:

A. In the Claude app (claude.ai) — easiest

  1. Sign in at claude.ai with your existing account (or create one — the free tier works for trying things out, though Fable lands on paid plans first).
  2. Open the model picker — the small model name shown near the message box. Click it.
  3. Choose "Fable 5." If it isn't listed yet, your plan's rollout day hasn't arrived; Opus 4.8 remains an excellent stand-in until it does.
  4. Ask it something real. Don't test with trivia — give it an actual task from your day: "summarize this document," "plan this project," "explain this bill." Capability differences show up on real work, not riddles.

B. In Claude Code — for website and coding work

  1. Open Claude Code in your terminal (or the desktop app). If you haven't installed it, the Claude Code docs walk through setup for Windows, Mac, and Linux — and our own OS guides cover the same in plainer language.
  2. Type /model and pick Fable 5 from the list. That's the whole switch — it also saves as your default for future sessions.
  3. Give it a long-horizon job. Fable's headline strength is working autonomously across millions of tokens — multi-file builds, migrations, overnight runs. The bigger and better-specified the task, the more the new model shows its worth. Stripe reported it "compressed months of engineering into days" on a codebase migration.

C. Through the API — for builders

  1. Get an API key from the Claude Platform console if you don't have one.
  2. Use the model ID claude-fable-5 — exactly that string, no date suffix.
  3. Mind the new rules: Fable 5 accepts adaptive thinking only (thinking: {"type": "adaptive"}). Sending the old budget_tokens style, sampling parameters (temperature, top_p, top_k), or even an explicit thinking: {"type": "disabled"} returns a 400 error — omit what you don't need.
  4. Budget deliberately. At $10 in / $50 out per million tokens, Fable costs 2× Opus 4.8. Route the hard, long-horizon work to Fable and keep everyday calls on Opus or Sonnet.

04Quick-reference card

ThingValue
Public model nameClaude Fable 5 ("Mythos-class")
API model IDclaude-fable-5
Restricted siblingClaude Mythos 5 — Glasswing partners; biology program in coming weeks
Context window1 million tokens (~2,500 pages in one conversation)
Max output128K tokens
Pricing$10 / M input · $50 / M output (2× Opus 4.8)
ThinkingAdaptive only — the model decides when and how hard to think
App rolloutPhased across plans, June 9–23, 2026
Safety classifiersFire in <5% of sessions; reroute narrow topics to Opus 4.8
AnnouncedJune 9, 2026

05Resources worth bookmarking

The official announcement Anthropic's own write-up — benchmarks, the science results, the safety case. The primary source everything on this page condenses. anthropic.com/news Claude developer docs Model overview, pricing tables, API reference, and migration notes — the authoritative spec for claude-fable-5. platform.claude.com/docs Claude Code docs Installing and driving the coding agent where Fable does its long-horizon best — setup, slash commands, skills, settings. code.claude.com/docs Anthropic status page Launch weeks mean load. If Fable seems slow or missing, check here before troubleshooting your own setup. status.anthropic.com

And on this site: the condensed announcement breakdown, the assistant comparison, and the device setup guides for getting Claude running on whatever machine is in front of you.

06Videos

07Courses & structured learning

Anthropic Academy Anthropic's free, self-paced courses — Claude fundamentals, prompting, building with the API, and Claude Code. The right starting point, straight from the source. anthropic.com/learn · free DeepLearning.AI short courses Andrew Ng's platform runs free 1–2 hour courses built with Anthropic on prompting Claude and building agents. Short, hands-on, beginner-friendly. deeplearning.ai · free

Neither has a Fable/Mythos-specific course yet — the announcement is a day old — but everything in them transfers directly: prompting, tool use, and agent patterns work the same on Fable 5, just better.

08Train in person: Austin

If you're in Austin and want immersion rather than self-paced study, the standout option sits in the downtown tech district, around the corner from Capital Factory:

09What people miss (so you don't)

Sources: Anthropic, "Claude Fable 5 and Claude Mythos 5" (June 9, 2026); Gauntlet AI; Capital Factory. Figures and quotes as reported in the official announcement. See also the condensed announcement breakdown — and from the archive, our May 2026 Mythos preview, written before launch.

Live
◐ Theme