Claude Fable 5,
and what it costs you.
Anthropic released Fable 5 on 9 June 2026 and described it, unusually plainly, as a Mythos-class model made safe for general use. The pricing tells you more about its intended job than the benchmarks do.
Fable 5 is the most capable model Anthropic has made generally available, with a one million token context window and up to 128,000 tokens of output. It is priced at $10 per million input tokens and $50 per million output. Both numbers are double what Opus 4.8 cost, which is the first useful thing to know, because it means the model is not a drop-in upgrade so much as a deliberate choice for specific work.
What the extra money is actually for
Anthropic points at software engineering, knowledge work, vision and scientific research as the areas where it leads. In practice, the case for paying twice as much is long-horizon work: problems where a slightly better decision at step three saves an hour at step thirty. On a short, well-specified task the cheaper model usually produces the same answer, and you have paid double for the privilege of reading it.
The safeguards are a behaviour, not a disclaimer
Fable 5 ships with classifiers covering cybersecurity, biology and chemistry, and attempts at distillation. Certain requests get routed to Claude Opus 4.8 instead. Anthropic is candid that these will sometimes catch harmless requests, and puts the rate at under 5% of sessions on average. If you work in security or life sciences, that is a number worth knowing before you build a workflow that assumes uninterrupted access.
The subscription window closed on 22 June
At launch, Fable 5 was included on the Pro, Max and Team plans at no extra cost through 22 June, after which usage credits are required. Anyone who tried it during that fortnight and built it into a routine will have felt the change. Worth checking what your team actually switched to when the free window closed, because the default is rarely a decision anyone made deliberately.
How to decide without guessing
Take a task you run regularly, run it on Fable 5 and on the cheaper model, and compare the outputs rather than the marketing. For a large share of everyday work the difference is invisible. For the work where it is visible, it tends to be very visible indeed, and those are exactly the tasks worth routing upward.
Where it fits alongside everything else
Treat the model line-up as a routing problem rather than a status symbol. Small and fast for volume, mid-range for most of the day, frontier for the handful of problems that genuinely reward it. Teams that pick one model for everything overpay on the routine work and underpower the hard work, usually at the same time.
Model prices, plan inclusions and access windows all changed inside the first fortnight of this model existing, so treat every figure here as a snapshot from June 2026 and verify against Anthropic’s current pricing before you commit a budget to it. If you want help working out which of your workloads justify a frontier model, that is a normal conversation in Claude Code coaching.
Things people
usually ask.
It was included at no extra cost on the paid plans for a limited window after launch, after which usage credits are needed. Plan inclusions change, so check what your own account currently offers rather than relying on launch-week coverage.
It refers to the underlying model family that Fable 5 and Mythos 5 share. Fable 5 is the version made safe for general availability; Mythos 5 is the restricted variant with some safeguards lifted, available only to approved organisations.
Almost certainly not. It costs double the previous flagship for input and output alike, and most day-to-day work does not produce a visibly better result on it. Route the hard problems to it and leave everything else where it is.