Anthropic's most powerful generally available model — a new tier above Opus, bringing Mythos-class capability to the public.
Fable 5, released June 9, 2026, is Anthropic's answer to a question they had been sitting on: how to ship the capabilities of Mythos — the internal frontier model that made headlines before release — in a form the public can actually buy. Fable 5 brings that capability tier to general availability, while the full Claude Mythos 5 stays in limited release through Project Glasswing. It is the first Claude model to sit above Opus in the lineup rather than replacing it.
At $10 input and $50 output per 1M tokens, it is the most expensive Claude by a factor of two — and Anthropic notes that's still half the price of the earlier Mythos Preview. The 1M context window finally gives the Claude family a flagship that matches GPT-5.5's window, and the 128K output ceiling is the largest Anthropic has shipped.
The plan story has a twist worth knowing before you subscribe for it. From launch (June 9) through June 22, 2026, Fable 5 is included at no extra cost on Claude Pro ($20/mo), Max ($100–200/mo), Team, and seat-based Enterprise plans. From June 23, Anthropic removes it from those plans' base allowance — using it then requires usage credits, pending capacity for re-inclusion. The API and consumption-based Enterprise plans have full access with no date cliff. If you want to evaluate it on a $20 plan, the free window is the time.
Anthropic describes Fable 5's capabilities as exceeding any model it has ever made generally available, with state-of-the-art results on nearly all tested benchmarks — software engineering, knowledge work, vision, and scientific research in particular. Early testers consistently call it a clear step beyond Opus 4.8 on complex, long-horizon reasoning and autonomous problem-solving: the multi-hour agentic tasks where prior models drift.
One unusual design choice: requests touching restricted high-risk areas — cybersecurity, biology and chemistry, model distillation — are automatically routed to Opus 4.8 instead, with a notice when that happens. In practice you get an answer from a still-frontier model rather than a refusal. The familiar Claude caveat also applies: no image, video, or audio generation.
| Model | Input / 1M | Output / 1M | Context |
|---|---|---|---|
| Claude Fable 5 | $10 | $50 | 1M |
| Claude Opus 4.8 | $5 | $25 | 200K |
| Claude Sonnet 4.6 | $3 | $15 | 200K (1M β) |
| GPT-5.5 pro | $30 | $180 | 1M |
| GPT-5.5 | $5 | $30 | 1M |
The pricing logic: Fable 5 costs 2× Opus 4.8 and roughly 2× GPT-5.5 — but sits at a third of GPT-5.5 pro's input rate and well under a third of its output rate, while claiming the stronger benchmark record. If your task runs fine on Opus 4.8 or Sonnet 4.6, stay there; reach for Fable 5 when the task is genuinely hard, long, or expensive to get wrong.