How to Access Claude Fable 5 and Mythos 5: Day-Zero Setup Guide (June 2026)

How to Access Claude Fable 5 and Mythos 5: Day-Zero Setup Guide (June 2026)

Anthropic shipped Claude Fable 5 on June 8, 2026 — the first publicly released model in the new Mythos-class family. Independent testing from Every's team puts it at 91/100 on the Senior Engineer benchmark, compared to Opus 4.8's 63 and GPT-5.5's 62. Anthropic calls it the most capable model it has ever made available to the public.

If you want to actually use it today — without waiting on a waitlist or reading three different changelogs — this guide walks you through every access path: Claude.ai, Claude Code, the desktop app, and the Anthropic API. We also cover the difference between Fable 5 (the consumer release) and Mythos 5 (the limited-access variant without safeguards), real pricing, and where Fable beats or wastes your money.

---

What Is Claude Fable 5?

Claude Fable 5 is Anthropic's June 2026 flagship model. It's the first public release of the "Mythos-class" architecture, a class of models Anthropic restricted in April when an earlier Mythos version was deemed too risky to release without guardrails. After months of red-teaming and partner feedback, Anthropic shipped Fable 5 as the public-safe variant: the same base model with safeguards that route cybersecurity and biology queries to Claude Opus 4.8 instead of answering them directly.

Three things make Fable 5 different from Opus 4.8 or Sonnet 4.6:

1. End-to-end task ownership. Fable 5 is built for long, async assignments — multi-hour runs where it plans, uses tools, repairs its own output, and ships a finished result. It's not designed for the back-and-forth rhythm of a chat.
2. Code quality at the senior level. On Every's Senior Engineer benchmark, Fable 5 scored 91/100 — near the range of human senior engineers who've taken the same test. Opus 4.8 scored 63; GPT-5.5 scored 62.
3. One-shot complex builds. In day-one testing, the Every team built a custom transcription app, a pixel-perfect 3D rendering of Borges's Library of Babel, and a strategic conversion-optimization report — each from a single prompt.

Fable 5 vs Mythos 5: which one are you actually getting?

This is the part most coverage glosses over. There are two models in the Mythos-class release:

For 99% of developers, Fable 5 is the version you'll be using — and for everything except the two restricted domains, the behavior is identical.

---

How to Access Claude Fable 5 (4 Paths)

Anthropic released Fable 5 simultaneously across four surfaces: the Claude web app, Claude Code, the desktop app, and the Anthropic API. Each one has trade-offs.

1. Claude.ai (Web App)

The fastest way to try Fable 5 with zero setup.

Steps:

1. Go to claude.ai and sign in to a paid Claude plan (Pro, Max, or Team).
2. Open a new chat. The model picker now includes "Claude Fable 5" alongside Opus 4.8 and Sonnet 4.6.
3. Select Claude Fable 5 from the dropdown.
4. Start prompting.

When this is the right surface: quick one-shot prompts where you want to read the output in the browser. Good for prototyping a single artifact (a web app, a research report, a code module) without setting up tooling.

Limitations: no local file access, no shell execution, no long async runs. You're paying per-token from your Claude subscription's allowance, which Fable 5 can burn through quickly.

2. Claude Code (CLI)

Claude Code is Anthropic's terminal-based coding agent. Fable 5 inside Claude Code is the most powerful way to use the model for engineering work, because it can plan, edit files, run tests, debug failures, and iterate over multi-hour runs.

Steps:

1. Install or update Claude Code:

   npm install -g @anthropic-ai/claude-code
   # or update an existing install:
   claude-code update
   

2. Authenticate (if you haven't already):

   claude-code login
   

3. Set Fable 5 as the active model for a session:

   claude-code --model claude-fable-5
   

   Or use the in-CLI /model command after launching:

   /model claude-fable-5
   

4. Start a session in your project root and give it a complete task — Fable 5 is designed to own the whole assignment.

When this is the right surface: large engineering tasks (build a feature end-to-end, refactor a module, port a library, write an integration with tests). Match Fable 5's design intent — async, long-running, fully framed.

Pricing note inside Claude Code: Fable 5 inside Claude Code counts against the same monthly token allowance as the API price tier ($10/M input, $50/M output). Heavy use will eat into your usage limit fast.

3. Claude Desktop App

The desktop app (Mac, Windows, Linux) gives you Claude.ai's chat experience plus local file integrations (artifacts, attachments) and a faster keyboard-first flow than the web.

Steps:

1. Download the latest Claude desktop app from claude.ai/download.
2. Sign in with your Claude account.
3. Open settings → Model preferences → select Claude Fable 5.
4. Start a chat. The desktop app supports drag-and-drop file attachments and a system tray quick-prompt.

When this is the right surface: writing tasks, document analysis, research synthesis — anywhere you want chat polish plus local file access but don't need shell execution.

4. Anthropic API

For programmatic access — building a product, an agent, a batch pipeline — the API is the only path.

Model ID: claude-fable-5 (confirm in the Anthropic API docs at release).

Minimal Python example:

import anthropic

client = anthropic.Anthropic()  # picks up ANTHROPIC_API_KEY from env

message = client.messages.create(
    model="claude-fable-5",
    max_tokens=4096,
    messages=[
        {
            "role": "user",
            "content": "Refactor this 200-line module into 4 testable units. Return a single diff."
        }
    ]
)

print(message.content[0].text)

Minimal TypeScript example:

import Anthropic from "@anthropic-ai/sdk";

const client = new Anthropic();

const message = await client.messages.create({
  model: "claude-fable-5",
  max_tokens: 4096,
  messages: [
    {
      role: "user",
      content: "Build a complete /pricing page in Next.js App Router that loads from a JSON config and matches our existing Tailwind theme. Return all files in a single response."
    }
  ]
});

console.log(message.content[0].text);

API-only features worth knowing:

• Prompt caching is supported on Fable 5 — critical when you're running multi-step agents. Cache your system prompt and tool definitions and you'll see meaningful cost savings on long runs.
• Tool use / function calling works the same as on Opus 4.8 — same schema, same response shape. Existing Claude tool implementations port over without code changes.
• Streaming is supported. Use it; Fable 5 runs are slow, and streaming gives you progress visibility instead of a 90-second blank wait.

---

Claude Fable 5 Pricing (and Why It Matters)

Fable 5 is roughly 2× the cost of Opus 4.8 and 3× the cost of Sonnet 4.6 on a per-token basis. Combined with its slower runs and higher token usage on complex tasks, a single long Fable 5 session can cost 5–10× what the same task would cost on Sonnet.

The practical implication: don't reach for Fable 5 by default. Save it for jobs where you'd otherwise hire a senior engineer for a half-day. For everything else, Sonnet 4.6 or Opus 4.8 is the right call.

---

When to Use Fable 5 vs Opus 4.8 vs Sonnet 4.6

The headline rule from Every's day-zero review: Fable 5 rewards a sharp problem frame and a strong review process. If you can't define the whole assignment before you hit send, you'll waste tokens on a model that isn't optimized for clarification rounds.

---

How to Get the Best Results from Fable 5

Three patterns from the first week of public use:

1. Treat it as an async agent, not a chat partner

Fable 5's design intent is "give it a complete task, walk away for 30–90 minutes, come back and review." The worst way to use it is to send a vague prompt and try to steer it interactively. The best way is to frame the whole assignment up front:

• Problem statement (one paragraph)
• Constraints (what NOT to do)
• Examples of the shape of the answer you want
• Success criteria (how you'll judge the output)

Send that, let it run, and review carefully.

2. Run several jobs in parallel

Because each Fable 5 run is slow, the rate-limiter is your review capacity, not the model. Kick off three or four independent runs at once on different problems. By the time you finish reviewing the first output, the others are done.

3. Invest in the framing, not the iteration

Most chat-style prompt engineering tips don't apply. You're not refining over 10 turns; you're writing one well-framed brief and reading one detailed output. The skill that matters is the same skill that makes you good at delegating to a senior contractor: write a tight, testable brief.

---

Common Setup Issues (and Fixes)

"Claude Fable 5" isn't in my model picker on Claude.ai You're either on a free plan or your account hasn't picked up the rollout yet. Sign in to a paid plan (Pro, Max, or Team) and reload. Anthropic gates new model rollouts behind paid tiers and rolls them out region by region over the first 24 hours.

Claude Code says "model not found: claude-fable-5" Your CLI is out of date. Run claude-code update or reinstall via npm install -g @anthropic-ai/claude-code. Anthropic ships model availability in CLI updates within hours of release.

API returns 404 on model: "claude-fable-5" Two possibilities:

1. Your API key is on an organization that hasn't been migrated to the new model yet — check the models endpoint for the exact model ID your org sees.
2. You're hitting an older API version. Make sure your client SDK is on the latest @anthropic-ai/sdk (Node) or anthropic (Python).

Fable 5 routes my question to Opus 4.8 unexpectedly If your question touches cybersecurity (offensive security, malware, exploit development) or biology (gain-of-function, synthesis routes), Fable's safeguards re-route to Opus 4.8 automatically. The response itself will note the routing. If you need un-routed access for legitimate research, you'd need Mythos 5 partner access — apply through Anthropic's enterprise contact.

I'm hitting usage limits within hours Fable 5 is token-hungry on complex tasks. A single multi-hour engineering run can easily consume 200K+ tokens. Check your monthly limit on the Claude billing page and consider upgrading to Max or Team if you're running Fable jobs daily.

---

What You Can Actually Build (Day-One Examples)

From Every's published day-zero test (seven testers, one week of use, single prompts each):

• Custom transcription app — a web app that brightens and levels grainy audio lectures, transcribes them, and lights up each sentence as the speaker says it. Built and deployed in one prompt.
• 3D Library of Babel — pixel-perfect rendering of Borges's hexagonal-room infinite library, fully navigable in browser. Single prompt.
• PULSE Browser Studio — a browser-based creative tool. Single prompt.
• Strategic conversion-optimization report — synthesized hundreds of survey responses + an entire web analytics stack, identified the biggest conversion issue, and proposed a clean falsifiable experiment.

The common pattern: one well-framed brief, one Fable 5 run, one finished artifact. Not "iterate over twenty turns until it works."

---

Should You Switch from Opus 4.8 to Fable 5?

For most developers, no — not as a default. The right move is to keep Opus 4.8 (or Sonnet 4.6) as your everyday driver and reach for Fable 5 deliberately when you have a long, complex, async-friendly assignment.

The cost math is brutal if you misuse it: Fable 5 at $50/M output tokens on a multi-hour run can cost $5–$20 per session. That's fine when the output is worth a half-day of senior engineering time. It's wasteful when you're asking a quick question.

The Every team's verdict — "a warp drive for power users, overpowered for everyone else" — is a fair summary. If you're already at the top of the AI adoption curve (using Claude Code in production, building agents, delegating real work), Fable 5 will unlock new categories of task. If you're earlier on that curve, you'll struggle to find tasks that justify the wait and the spend.

---

Bottom Line: Day-Zero Action Plan

If you want to use Fable 5 today, here's the shortest path:

1. Sign in to your paid Claude plan at claude.ai — verify Fable 5 is in your model picker.
2. Update Claude Code with claude-code update or a fresh npm install.
3. Pick one big, well-framed task that's been sitting in your backlog because it would take half a day of senior engineering. Write a complete brief.
4. Run it on Fable 5 via Claude Code (--model claude-fable-5) and walk away.
5. Review carefully when it finishes. The model's ceiling is high, but the responsibility for judging quality stays with you.

For everything else, keep Sonnet 4.6 or Opus 4.8 on hand. Match the model to the assignment.

---

Working with Claude Fable 5 in Production?

If you're shipping AI features into production — agents, multi-step pipelines, Claude Code workflows — Fable 5's pricing and pace make engineering judgment more valuable, not less. The teams that get the most out of it pair the model with senior AI engineers who know how to frame the assignment and review the output.

AY Automate places senior AI engineers (Claude Code, agents, MCP servers, RAG, evaluation pipelines) into your team for 30–90 day engagements. If you'd like a 30-minute call to map what Fable 5 unlocks for your team, book a free strategy call — Walid runs them personally, no slides, no sales pitch.