Claude Fable 5 Explained: Anthropic's Mythos-Class Model
Anthropic built a model strong enough that the government paused it. Then it went on sale.
Claude Fable 5 is the most capable model Anthropic has ever made generally available — a new Mythos-class tier that sits above Claude Opus in the lineup, launched June 9, 2026. Its first month was eventful even by 2026 standards: state-of-the-art results across nearly every benchmark, a US export-control order that restricted access within days of launch, and a full global release on July 1 once the controls were lifted. If you have been seeing the names Fable and Mythos everywhere and wondering what is actually going on, this is the plain-language version.
- Claude Fable 5 launched June 9, 2026 as a new Mythos-class tier above Opus 4.8.
- Fable 5 and Mythos 5 are the same underlying model — Fable adds safety measures for general availability; Mythos goes only to approved organizations.
- US export controls briefly restricted access in mid-June; they were lifted June 30 and Fable 5 went global July 1.
- API pricing is $10 per million input tokens and $50 per million output.
- It currently tops independent capability rankings — ahead of Opus 4.8, GPT-5.5, and Grok 4.5.
What is Claude Fable 5, exactly?
For years Anthropic's ladder was simple: Haiku for speed, Sonnet for balance, Opus for power. Fable 5 introduces a fourth rung — a Mythos-class model whose capabilities exceed anything the company had previously shipped publicly. Anthropic reports state-of-the-art performance on nearly all tested benchmarks, with standout results in software engineering, knowledge work, vision, and scientific research, and independent rankings agree: on the Artificial Analysis Intelligence Index, Fable 5 scores 60, ahead of Opus 4.8 at 56, GPT-5.5 at 55, and the newly released Grok 4.5 at 54.
In everyday use, the difference shows up on the tasks where lesser models plateau: multi-hour engineering problems, research synthesis across dozens of sources, and the kind of ambiguous, multi-constraint planning that used to require breaking a problem into a dozen carefully supervised steps.
Fable vs Mythos: what's the difference?
They are the same underlying model with different deployment rules — the split is about who gets which safeguards, not two different brains:
| Claude Fable 5 | Claude Mythos 5 | |
|---|---|---|
| Who can use it | Everyone (generally available) | Approved organizations only — initially cyberdefenders and infrastructure providers |
| Safeguards | Built-in safety measures for dual-use domains; sensitive queries are answered by Claude Opus 4.8 instead | Safeguards lifted in specific areas for defensive work |
| Notable | State-of-the-art on public benchmarks | Deployed via Project Glasswing with the US government; strongest cybersecurity capabilities of any model |
The Fable safeguard design is genuinely novel: rather than refusing outright, queries that touch sensitive territory — certain cybersecurity, biology, and chemistry topics — quietly fall back to a response from Opus 4.8. For the overwhelming majority of work you will never encounter the seam. Anthropic's full explanation is on its announcement page.
Why was Claude Fable 5 restricted, then released globally?
Three days after launch, on June 12, the US government applied export controls to Fable 5 and Mythos 5 — the first time a commercial AI model was treated this way — requiring Anthropic to restrict access for foreign nationals while the models were evaluated. On June 30 the controls were lifted, and starting July 1 Fable 5 became available to users globally across the Claude apps and API.
Whatever your politics on AI regulation, the episode is a milestone: frontier models are now capable enough that governments treat their release the way they treat advanced hardware. It is also, frankly, the strongest endorsement a benchmark chart could never give.
What does Claude Fable 5 cost — and when is it worth it?
API pricing is $10 per million input tokens and $50 per million output — less than half the price of the earlier Mythos Preview, but still premium territory: five times Sonnet 5's intro input price, double GPT-5.6 Sol's. So the practical question is not "is it the best model" (it is, by most measures) but "which of my tasks deserve it."
A sane routing rule: use Claude Sonnet 5 or GPT-5.6 Terra as your daily default, and escalate to Fable 5 when the task is genuinely hard — the architecture decision, the contract review, the research question where a subtly wrong answer costs real money. On CoreAI, escalating is one tap: switch the model mid-conversation in the web app, or put Fable 5 head-to-head against your current favorite in Compare and see whether the delta shows up on your own prompts. Every plan's budget works across all 300+ models in the library, Fable 5 included — no separate frontier-tier subscription.
What can you actually do with a Mythos-class model?
Benchmarks are abstract; workloads are not. In practice, Claude Fable 5 earns its premium on the tasks that used to require a human expert in the loop: auditing a 20-file pull request and catching the one interaction that breaks production, reading a full contract stack and flagging the clause that conflicts with the term sheet, or synthesizing thirty research papers into a defensible recommendation rather than a book report. Vision work takes a similar jump — dense charts, handwritten notes, and messy PDFs that trip mid-tier models parse cleanly. The pattern across all of it: fewer retries, fewer hallucinated details, and answers that survive scrutiny from someone who knows the domain. That reliability, more than any single benchmark point, is what a Mythos-class tier buys you — the model fails less often precisely on the tasks where failure is expensive.
Frequently Asked Questions
What is the difference between Claude Fable 5 and Claude Mythos 5?
Same underlying model, different deployments. Fable 5 is generally available and includes safety measures — sensitive dual-use queries receive a response from Opus 4.8 instead. Mythos 5 has those safeguards lifted in specific areas and is restricted to approved organizations, initially cyberdefenders working with the US government under Project Glasswing.
Is Claude Fable 5 the most powerful AI model right now?
By most public measures, yes. It leads the Artificial Analysis Intelligence Index at 60 — ahead of Opus 4.8 (56), GPT-5.5 (55), and Grok 4.5 (54) — and Anthropic reports state-of-the-art results on nearly all benchmarks it tested.
How much does Claude Fable 5 cost?
$10 per million input tokens and $50 per million output via the API — less than half the price of the earlier Mythos Preview. On CoreAI it is included in the standard plans, sharing the same usage budget as every other model.
Why did the US government put export controls on Fable 5?
On June 12, 2026, export controls required Anthropic to restrict access for foreign nationals while the model's advanced capabilities — particularly in cybersecurity — were assessed. The controls were lifted June 30, and Fable 5 became globally available July 1.
Can I try Claude Fable 5 without a Claude subscription?
Yes. Claude Fable 5 is available on CoreAI's web, iOS, and Android apps under one subscription that also includes GPT-5.6, Gemini, Grok, DeepSeek, and 300+ other models.
Put Fable 5 to work on CoreAI
The strongest public model, next to every rival worth testing — one subscription, 300+ models.