Claude Fable 5 & Mythos 5: Anthropic's New Model Class Explained

What is Claude Fable 5?

Claude Fable 5 is Anthropic's new flagship model, introduced on June 9, 2026. It establishes the "Mythos class" — a new capability tier above the previous Opus models — and is generally available right away: via the Claude API and on Claude's paid plans.

The key specs according to Anthropic's documentation: API name claude-fable-5, a context window of 1 million tokens (enough for entire codebases or hundreds of pages of documents in one go), and up to 128,000 output tokens. On the Pro, Max, Team, and Enterprise plans, Fable 5 is included at no extra cost from June 9 to 22; after that, usage runs on credits.

The naming break is notable: for years the tiers were called Haiku, Sonnet, Opus — now Anthropic adds an entire class on top with Fable and Mythos. For how the established models from OpenAI, Google & Co. perform in everyday work, see our AI tool comparison 2026.

"Fable 5 opened up a class of long-horizon problems that were out of reach." — Michael Truell, CEO of Cursor, in the Anthropic announcement

What is Claude Mythos 5 — and why can't everyone use it?

Claude Mythos 5 is the same base model as Fable 5 — with some safeguards lifted. Access is limited to vetted partners: currently security teams from Anthropic's "Project Glasswing" (cyber safeguards lifted), soon selected biology researchers. For everyone else, Mythos 5 stays locked — deliberately.

The background: Fable 5 runs with three protective classifiers — against offensive cyber use, against biology and chemistry risks, and against "distillation", i.e. mass-harvesting its answers to train competing models. When a classifier triggers, the previous model Claude Opus 4.8 takes over automatically. According to Anthropic, this affects less than 5% of all sessions; an external bug bounty found no universal jailbreak in over 1,000 hours of testing. All Mythos-class traffic is also subject to a 30-day retention requirement — with no use for model training.

What the less-restricted variant delivers in researchers' hands is already hinted at in the announcement: Mythos 5 produced strong drug-design candidates for 9 of 14 target proteins, and in blind comparisons scientists preferred its molecular-biology hypotheses about 80% of the time over the Opus class. In genomics, the model processed single-cell data from millions of cells across 138 animal species — and built its own ML model for the job that beats a recent Science publication despite being 100 times smaller.

How good is Claude Fable 5 really?

Based on everything Anthropic and early partners are showing: the strongest publicly available AI model right now — especially on long, multi-step tasks. Fable 5 leads several coding and finance benchmarks and is the first model to break the 90% mark on the core analytics benchmark. One honest caveat belongs here: nearly all of these numbers come from Anthropic's own announcement; independent testing is still ahead.

All figures from the Anthropic announcement of June 9, 2026. Real-world voices add to it: Stripe says Fable 5 "compressed months of engineering into days", GitHub's Chief Product Officer Mario Rodriguez attests to autonomy and reliability "exceeding previous benchmarks", and Anthropic's own research note rates its reasoning as "senior research scientist grade".

How much does Claude Fable 5 cost?

$10 per million input tokens and $50 per million output tokens — making Fable 5 twice as expensive as Claude Opus 4.8 and by far the priciest model in the Claude lineup. In return, Anthropic promises higher token efficiency: Fable 5 uses fewer tokens than its predecessors for the same task.

Sources: Anthropic announcement and model documentation (as of June 10, 2026). Worth knowing for subscribers: Fable 5 is included in paid plans at no extra cost until June 22; after that it runs on usage credits until Anthropic has enough capacity to restore it to standard plans.

What does the launch mean for small and medium businesses?

In short: better agents, same homework. Fable 5 makes AI agents more persistent and reliable on long task chains — exactly where things used to break down. But the ground rule doesn't change: first comes a clean process, then the model. And for routine work, a cheaper model remains the better deal.

The biggest lever is autonomous agents: according to Anthropic, both models work autonomously for longer than any previous Claude model — including their own file-based memory across sessions. Which processes such agents already handle in SMBs today, and how to start safely, is covered in our guide to AI agents for SMBs.

Second: model choice is a cost question. For emails, copy, and standard automations, Sonnet or Haiku are enough — at a fifth to a tenth of the price. Fable 5 pays off where tasks used to fail: long research runs, code migrations, multi-step analyses. For a neutral overview of all the major providers, see our AI tool comparison.

Third: the pace stays high — just last week it was Anthropic's IPO plans, now it's a whole new model class. If you build your automations modularly, for instance with n8n, you swap the model later via a dropdown instead of a rebuild. That's exactly how we build in client projects — see our services and real results on our references page.