Claude Fable 5 explained: the first AI above Opus you can actually use.

On June 9, 2026, Anthropic released Claude Fable 5 — the first publicly available model in a new tier that sits above Opus. The headlines say "most powerful model ever." Here's what that actually means for your work, in plain English, plus the catch most coverage skips: free access ends June 22.

The short version

Anthropic's models come in tiers. Until now the top tier you could use was Opus. In April, Anthropic quietly built a tier above it — internally called Mythos-class — but kept it locked to a small group of cybersecurity partners because it was considered too capable to release widely.

Fable 5 is that Mythos-class model, fitted with safety guardrails so the general public can use it. Same underlying brain as the restricted version; the guardrails are the only difference. So when you hear "Fable 5 is the most powerful model Anthropic has ever made generally available" — that's literally true, and it's a meaningfully bigger jump than the usual version bump.

01 Why there are two names: Fable vs Mythos

This confuses everyone, so let's settle it. Fable 5 and Mythos 5 are the same model. The names describe the safeguards, not the capability:

• Claude Fable 5 — the public version, available everywhere today. It has safety classifiers that, on certain sensitive topics, hand your request to Claude Opus 4.8 instead of answering directly. This happens in fewer than 5% of sessions.
• Claude Mythos 5 — the same model with some of those safeguards lifted. It is not public: it's restricted to vetted cyber-defense partners (through a government program called Project Glasswing) and, soon, a small set of biology researchers.

Anthropic even named them deliberately: "Fable" comes from the Latin fabula, "that which is told," a cousin of the Greek mythos. The story is the same; what you're allowed to hear differs. For 99% of people, Fable 5 is simply "the new top Claude," and Mythos is a name you'll see in headlines but never touch.

02 What actually got better

Benchmarks are boring. Here's what the capability jump looks like in real terms, drawn from Anthropic's launch testing:

It works on its own, for much longer

The single biggest change: Fable 5 can work autonomously on long, multi-step tasks far longer than earlier models without losing the thread. Anthropic's line is "the longer and more complex the task, the larger Fable's lead." The headline example: the payments company Stripe reported Fable 5 did a codebase-wide migration in a day that would have taken a team over two months by hand.

It sees better

Fable 5 is now state-of-the-art at vision tasks — reading precise numbers off a complex chart, or rebuilding a web app's code from nothing but a screenshot. If your work involves pulling information out of images, PDFs, dashboards, or scanned documents, this is the practical upgrade you'll feel.

It remembers and improves its own work

On long tasks, Fable 5 keeps notes for itself and uses them to get better as it goes — closer to how a person works through a big project over days than how a chatbot answers a single question.

03 The honest part: safeguards and your data

Two things changed that vendors tend to bury, and that you should know as a user.

Some requests get answered by a different model. Because a Mythos-class model is genuinely dangerous in the wrong hands (Anthropic specifically worries about cybersecurity and biology misuse), Fable 5 runs requests through safety classifiers first. If your request touches cyber, biology/chemistry, or a few other flagged areas, it's quietly answered by Claude Opus 4.8 instead — and you're told when that happens. Anthropic tuned these deliberately broad to ship fast, so they sometimes catch harmless requests. If you ever get a "this was handled by Opus 4.8" notice on an innocent question, that's why — not a bug, a conservative filter.

Business data is retained for 30 days. For Mythos-class models, Anthropic now requires 30-day retention of traffic (including business traffic) to help detect novel attacks and jailbreaks. They say they won't train on it or use it for anything beyond safety, and it's deleted after 30 days. If your company has an AI-use policy, this is a line worth checking before you run sensitive material through Fable 5. We dig into what this means for teams in Lesson 18.

04 What you should do this week

1. If you pay for Claude: open it, confirm you can select Fable 5, and run it on the single hardest, longest task you've got — the report, the analysis, the messy spreadsheet. The free window through June 22 is the cheapest time you'll ever get to judge whether the jump is real for your work.
2. If you're on the free Claude plan: you won't get Fable 5 directly, but the patterns still matter — and our free AI Foundations track teaches the prompting skills that make any model better.
3. Either way: notice the shift this model represents. The skill that matters now isn't writing a clever one-line prompt — it's handing off a whole task well. That's the next lesson.