Claude Fable 5 and Mythos 5 Explained: Anthropic's New AI in Plain English

Claude Fable 5 was Anthropic's top public model until Claude Fable 5.1 replaced it on September 1, 2026, on exactly the same plan terms -- and it had the strangest launch month in AI history. Anthropic released it on June 9, 2026 alongside a restricted twin called Claude Mythos 5. Three days later the US government slapped export controls on both models and they vanished. On June 30 the controls were lifted, and Fable 5 came back worldwide on July 1. If you subscribe to a paid Claude plan you have it today -- but as of July 20, how you pay for it depends on your plan. Free users do not get it at all.

What this means for you, in one line: if you are on Max, nothing changes -- keep using Fable 5 as before. If you are on Pro or a standard Team seat, Fable 5 still works, but it now bills against usage credits instead of your plan. To cushion that, Anthropic handed eligible Pro and Team standard seats a one-time $100 credit. That claim window closed on August 2, 2026, and every claimed balance expired on September 17, 2026 at 11:59 PM PT, so the cushion is gone for everyone.

Key Takeaways

• Fable 5 and Mythos 5 are the same model. The difference is safeguards: Fable 5 blocks high-risk cybersecurity and dual-use biology requests; Mythos 5 lifts some of those blocks for vetted partners only. Since August 6, 2026 the biology filter fires about 85% less often.
• Anthropic says Fable 5's capabilities "exceed those of any model we've ever made generally available," with state-of-the-art results in coding, finance, vision, and long-document work.
• The free-inclusion promo ended July 19, 2026. As of July 20, only Max plans and premium Team and Enterprise seats keep Fable 5 included (up to 50% of weekly limits). Pro plans and standard Team seats now pay per use with usage credits.
• The one-time $100 cushion credit is over. Anthropic gave eligible Pro and standard Team seats $100 of usage credits ($100 per purchased standard seat on Team, pooled, up to $2,500 an org), but the claim window shut on August 2, 2026 at 11:59 PM PT. Credits already claimed expired on September 17, 2026 at 11:59 PM PT.
• It is expensive machinery: $10 per million input tokens and $50 per million output tokens on the API -- the priciest model Anthropic sells broadly.
• The June export-control saga (applied June 12, lifted June 30) was a first for a commercial AI model and a preview of how frontier releases now work.
• One real-world data point: Stripe says Fable 5 compressed a code migration that would have taken a team two months into a single day.

What Are Fable 5 and Mythos 5?

They are one model wearing two different safety configurations. Anthropic built a system so capable in sensitive areas -- especially cybersecurity and biology -- that it decided not to ship it raw to the public. Fable 5 is the general-release version: the full model, plus classifier-based guardrails that block offensive hacking requests, route blocked biology and chemistry questions to a different model, and prevent competitors from extracting its capabilities. As of July 24, 2026 those biology questions land on Opus 5 rather than the older Opus 4.8, so the fallback is a much stronger model than it was -- and since August 6, 2026 far fewer questions are routed at all, after Anthropic retrained the biology classifier and cut those fallbacks by about 85%. Mythos 5 is the same model with some guardrails lifted, available only to vetted organizations: cyberdefense teams in Anthropic's Project Glasswing program and, increasingly, select biomedical researchers. Anthropic is blunt that "the safeguards are what distinguish the two models." For everyone reading this, Mythos 5 is a headline, not a product -- Fable 5 is the one that concerns you.

What Can Fable 5 Do That Older Claudes Couldn't?

The step up is about sustained, professional-grade work rather than smarter small talk. The most striking claim in Anthropic's announcement comes from Stripe, which reports Fable 5 completed a 50-million-line code migration in one day -- work previously scoped at two months for an engineering team. Beyond coding, Anthropic reports the highest score of any model on Hebbia's finance benchmark for senior-analyst reasoning, new state-of-the-art vision performance (reading numbers off charts, rebuilding an app from a screenshot), and a 3x improvement over Opus 4.8 on memory-intensive tasks, staying coherent "across millions of tokens" -- meaning book-length projects no longer fall apart halfway through.

What it deliberately cannot do: help with offensive hacking (Anthropic reports zero harmful responses across tested jailbreak techniques, plus a post-incident classifier that blocks a newly discovered jailbreak in over 99% of cases), and dual-use biology -- virology, toxicology and molecular design -- which quietly falls back to another model, since July 24, 2026 Opus 5. Anthropic says Fable 5 "isn't yet usable for professional biology research and drug development." Everyday health and educational biology questions are a different matter after the August 6, 2026 classifier update, which cut biology fallbacks by about 85%.

The Export-Control Saga, Briefly

For 19 days, the US government treated an AI model like a controlled weapon -- that is the part of this story worth remembering. On June 12, export controls were applied to both models, restricting access by foreign nationals; since Anthropic could not verify nationality in real time, it suspended access entirely. Mythos 5 was partially restored to US organizations on June 26, the controls were lifted June 30, and Fable 5 redeployed globally on July 1. You do not need to do anything about any of this, but it explains why your Claude app lost its best model for most of June -- and it set the template OpenAI is now following with its government-coordinated GPT-5.6 rollout.

Do You Get It, and What Does It Cost?

Only if you pay, and since July 20 your plan decides whether Fable 5 is included or metered. Here is the picture as of July 20, 2026:

Two details people get wrong. Fable 5 eats your allowance faster than other Claude models -- Anthropic says it "draws from your plan's regular weekly usage limit and uses it faster," so on Max the 50% cap can arrive sooner than you expect. And hitting that cap is not a wall: you either switch to another Claude model and keep working inside your normal limits, or turn on usage credits and pay separately. If credits are already on, billing switches over automatically once you pass the limit -- worth knowing before you keep hammering Fable 5 out of habit.

The one-time credit closed on August 2, 2026 (11:59 PM PT) and can no longer be claimed. It was $100 of standard usage credits ($100 per purchased standard seat on Team, pooled up to $2,500), usable on any model, not just Fable 5, and eligibility ran to anyone holding a qualifying Pro or standard Team seat as of July 19. Every claimed balance expired on September 17, 2026 at 11:59 PM PT, regardless of when it was claimed, and there is no second window: Anthropic's help article says that after the window closes "the credits can no longer be claimed." For context, the $10/$50 API price is steep -- Google's Gemini 3.1 Pro costs a fraction of that -- which tells you Anthropic is positioning this as a professional tool, not a casual chatbot.

Should You Care?

If you use AI for real work, yes -- this is the current capability ceiling. If you use AI to draft emails and settle dinner-table debates, mostly no.

Use Fable 5 if you are already a paying Claude subscriber doing substantial work: coding, long documents, financial analysis, or anything that previously fell apart in long sessions. If you are on Max, it is still included -- use up to 50% of your weekly limit on it at no extra cost. If you are on Pro or a standard Team seat, the one-time $100 credit expired on September 17, so any Fable 5 use now comes out of usage credits you pay for; test it on a small task first. Stick with your current setup if you are a free user or casual chatter -- Claude's included models, Gemini's free tier, and ChatGPT remain excellent for everyday use, and our guide to which AI chatbot deserves your money covers the tradeoffs.

And if the Stripe two-months-into-one-day number makes you nervous about your own job, that is a reasonable reaction with a more nuanced answer than the headline suggests -- we dig into it in which jobs AI can't replace in 2026.

The bottom line: Fable was the strongest AI model ordinary customers could buy, wrapped in the heaviest safety apparatus ever shipped with one, and Fable 5.1 took that position on September 1, 2026 without changing a single plan term. Max subscribers keep it included; Pro and standard Team users now meter it against usage credits, and the $100 cushion credit that softened the switch closed to new claims on August 2, 2026 and expired on September 17. Everyone else loses nothing by waiting for the price and access to settle.

Corrections and Updates

September 8, 2026: this page called Fable 5 Anthropic's most capable public model, and it stopped being that a week ago. Anthropic's help-centre release notes carry a September 1, 2026 entry, "Claude Fable 5.1 and Claude Mythos 5.1 launch", pointing at its launch post. The plan terms this page describes are unchanged by that release: still no Fable on Free, still usage credits on Pro, still 50% of weekly limits on Max, re-decoded from claude.com/pricing on September 8, 2026. The summary, lede and closing verdict were corrected; the rest of the page's record of the promotion, the export controls and the $100 credit stands. What 5.1 changes is in Claude Fable 5.1, explained.

August 7, 2026: the biology guardrail described on this page is now much narrower.

August 3, 2026: the one-time $100 usage credit is closed.

July 24, 2026: Anthropic released Claude Opus 5, which it says comes "close to the frontier intelligence of Claude Fable 5 at half the price" -- and unlike Fable 5, it is included on Claude Pro at no extra cost.

July 20, 2026: the promotion that included Fable 5 inside your plan's weekly usage limits ended July 19, 2026 at 11:59:59 PM PT -- and this time it landed rather than slipping again (it had moved twice before, from July 7 to July 12 to July 19).