Chapter 80·Beginner·10 min read
What Is Claude Fable 5? Anthropic's Mythos-Class Model Explained
What is Claude Fable 5, really? A plain-English explanation of Anthropic's most capable model — the new Mythos-class tier above Opus, how Fable 5 differs from Mythos 5, what it costs, and why it launched with the strongest safeguards Anthropic has ever shipped.
July 9, 2026
If you've been anywhere near AI news since June 2026, you've heard the name Claude Fable 5 — usually next to words like "most capable model ever released," and then, a few days later, next to words like "US government shutdown."
This guide covers the whole story in plain English: what Fable 5 actually is, what it can do, the safeguards it ships with, why the US government blocked it three days after launch, and how it came back. We start with the basics.
The simplest accurate definition
Claude Fable 5 is Anthropic's most capable widely released AI model — the first model in the Claude 5 family, and the first in a new Mythos-class tier that sits above the Opus tier that previously topped the range.
In Anthropic's own words from the launch announcement:
"Fable 5's capabilities exceed those of any model we've ever made generally available."
That sentence is doing careful work. Fable 5 is the most capable model you can get; Anthropic considers the underlying capability significant enough that it ships with the strongest safeguards the company has ever applied — and a nearly identical sibling without those safeguards exists for a much smaller audience.
Fable 5 vs Mythos 5: two doors to one model
Anthropic released the same underlying model under two names on June 9, 2026:
| Claude Fable 5 | Claude Mythos 5 | |
|---|---|---|
| Underlying model | Identical | Identical |
| Safety classifiers | Yes — can decline certain requests | No |
| Who can use it | Everyone (generally available) | Approved organisations in Project Glasswing only |
| API model ID | claude-fable-5 | claude-mythos-5 |
| Pricing | $10 / $50 per million tokens (in/out) | Same |
Mythos 5 succeeds the earlier invitation-only Claude Mythos Preview, at less than half its price. For almost everyone, "Fable 5" is the model that matters — and per Anthropic, more than 95% of Fable sessions never touch a safety fallback at all.
The spec sheet, translated
The numbers, and what they mean in practice:
| Spec | Value | In plain English |
|---|---|---|
| Context window | 1M tokens (default) | Roughly 700,000+ words of input — entire codebases or document sets in one request |
| Max output | 128K tokens per request | Long reports, large diffs, full documents in one go |
| Pricing | $10 in / $50 out per million tokens | Double Claude Opus 4.8 — this is the premium tier |
| Thinking | Always on (adaptive) | The model always reasons before answering; you tune depth via an effort setting |
| Data retention | 30 days, required | Mythos-class traffic is kept for safety review; zero-data-retention setups can't use it |
Two of those rows deserve unpacking.
Thinking is always on. On earlier Claude models, extended reasoning was something you switched on. On Fable 5 it's the only mode: the model decides how hard to think per request, and you steer it with an effort control rather than a token budget. Notably, the raw chain of thought is never returned — you can request a readable summary of the reasoning, but the verbatim internal reasoning stays private to the model.
The retention rule is a safety measure. Anthropic requires 30-day retention for all Mythos-class traffic so that safety teams can investigate misuse. The company states the data "won't be used… for any non-safety-related purpose," with human access logged.
Where it fits in the Claude line-up
Fable 5 isn't meant to replace Opus, Sonnet, or Haiku for everyday work — at 2× Opus pricing, it's aimed at the tasks those models can't do: multi-hour autonomous runs, million-token reasoning jobs, frontier-grade coding and research. We look at exactly what that means in the capabilities chapter.
A launch unlike any other
Fable 5's first month is a story in itself, and it's why this guide has two whole chapters on policy rather than technology:
- Jun 9, 2026Launch
Fable 5 and Mythos 5 launch — API, claude.ai, Claude Code.
- Jun 12, 2026Blocked
US government export-control directive; Anthropic disables both models for all users worldwide.
- Jun 30, 2026Controls lifted
Export controls lifted after investigation and new safeguards.
- Jul 1, 2026Restored
Fable 5 available again, globally.
A frontier AI model being switched off by a government order — and switched back on three weeks later — had never happened before. We cover why it was blocked and how it came back in dedicated chapters.
Frequently asked questions
Is Claude Fable 5 available right now?
Yes. After a three-week government-ordered suspension in June 2026, Fable 5 has been available globally since July 1, 2026 — on claude.ai (Pro, Max, Team, and select Enterprise plans), Claude Code, Claude Cowork, and the Claude API as claude-fable-5.
What's the difference between Claude Fable 5 and Claude Opus 4.8?
Fable 5 sits a full tier above Opus 4.8 — it's a Mythos-class model built for long-horizon agentic work and frontier reasoning, at double the price ($10/$50 vs $5/$25 per million tokens). Opus 4.8 remains the better value for everyday hard tasks; Fable 5 is for the jobs Opus can't finish.
Is Claude Fable 5 the same as Claude Mythos 5?
Same underlying model, different packaging. Fable 5 includes safety classifiers that can decline certain requests and is available to everyone. Mythos 5 removes those classifiers and is restricted to organisations approved for Project Glasswing.
How much does Claude Fable 5 cost?
$10 per million input tokens and $50 per million output tokens on the API. On claude.ai subscriptions it's included up to 50% of weekly usage limits through July 7, 2026, then draws usage credits — with Anthropic aiming to fold it back into standard plan limits.
Why was Claude Fable 5 taken offline in June 2026?
A US government export-control directive, issued June 12, 2026 after a jailbreak report, forced Anthropic to disable it worldwide. The controls were lifted June 30 after investigation showed the jailbreak exposed no unique capability. The full story is in why Fable 5 was blocked.
Recap
- Claude Fable 5 is Anthropic's most capable widely released model — first of the Claude 5 family, in a new Mythos-class tier above Opus.
- Fable 5 and Mythos 5 are the same model: Fable ships with safety classifiers; Mythos 5 removes them and is restricted to Project Glasswing partners.
- Specs: 1M-token context window, 128K output tokens, $10/$50 per million tokens, always-on adaptive thinking, mandatory 30-day data retention.
- The raw chain of thought is never returned — only optional summaries.
- Launched June 9, 2026; blocked by a US export-control order on June 12; restored July 1 — a first for the industry.
Next: what all that capability actually looks like in practice. Continue to Claude Fable 5's capabilities and benchmarks.