Reviews

Analysis: Claude Fable 5 — what's new, what it costs, and who it's for

Anthropic has released Claude Fable 5, its first generally available 'Mythos-class' model, at $10/$50 per million tokens. We look at the specs, the safeguard design that routes some requests to another model, and who it suits.

Abstract illustration of interweaving ribbons of light, in violet and magenta.
Abstract illustration of interweaving ribbons of light, in violet and magenta.

Anthropic released Claude Fable 5 on 9 June 2026, describing it in its announcement as “a Mythos-class model that we’ve made safe for general use.” The model is available through the Claude API under the ID claude-fable-5 and, for a limited period, on most paid Claude subscriptions. A companion model, Claude Mythos 5, is built on the same underlying model but is restricted to a small group of partners.

This analysis is based on Anthropic’s published announcement and documentation, and on benchmark results reported by Anthropic and its customers; we have not tested the model hands-on.

Update, 17 November 2026: On 12 June, two days after this analysis was published, Anthropic disabled access to Claude Fable 5 and Claude Mythos 5 for all customers, saying it was complying with a US government export-control directive. The company said publicly that it disagreed with the directive. Anthropic restored access to Fable 5 on 1 July, and has since released Claude Fable 5.1 (1 September) and Claude Fable 5.2 (9 November). The details below reflect Anthropic’s launch documentation as of 10 June.

What “Mythos-class” means

Anthropic uses the term for a tier above its Opus models. In the announcement’s own definition, “Mythos-class models are a tier of Claude models that sit above our Opus class in capability.” The first of these, Claude Mythos Preview, was released in April only to a limited group of cyber defenders and critical software infrastructure providers through a programme Anthropic calls Project Glasswing.

Fable 5 is the first model in that tier that Anthropic has opened to the general market, and the company says its capabilities exceed those of any model it has previously made generally available. Claude Mythos 5 is the same model with some safeguards lifted. Anthropic says it will first be deployed through Project Glasswing, in collaboration with the US government, as an upgrade to Mythos Preview, with plans to add selected biology researchers and, later, a broader trusted-access programme.

How the safeguards work

The most distinctive part of the launch is how Anthropic has packaged the model. Classifiers watch for requests related to cybersecurity, biology and chemistry, and distillation (using a model’s outputs to train another model). When one is detected, the response is handled by Claude Opus 4.8, which Anthropic calls its next-most-capable model, and the user is told that this has happened.

Anthropic says these safeguards trigger, on average, in less than 5% of sessions. It is also candid that the biology and chemistry filters are deliberately broad: the announcement says Fable will fall back “on most requests related to biology and chemistry”, and that the company aims to narrow this over time.

On robustness, Anthropic reports that an external bug bounty of more than 1,000 hours found no universal jailbreak, while noting that the UK AI Security Institute “made progress towards one within a brief initial testing window.”

There is also a data-handling condition. Anthropic will require 30-day retention for all Mythos-class traffic, including on third-party platforms. It states: “We won’t use this data to train new Claude models, or for any non-safety-related purpose.”

Specs and pricing

Claude Fable 5
API model ID claude-fable-5
Input price $10 per million tokens
Output price $50 per million tokens
Context window 1M tokens
Maximum output 128K tokens
Inputs and outputs Text and images in, text out
Thinking Adaptive, always on (default effort: high)
Reliable knowledge cutoff January 2026
Batch API 50% discount

Anthropic says the price is “less than half the price of Claude Mythos Preview”, and Mythos 5 carries the same rates. For context, Anthropic lists Claude Opus 4.8, the fallback model, at $5 input and $25 output per million tokens, so Fable 5 costs twice as much per token as the Opus tier.

Availability

  • Claude API: available from launch.
  • Claude subscriptions: from launch through 22 June, Fable 5 is included on Pro, Max, Team and seat-based Enterprise plans at no extra cost. From 23 June, using it on those plans will require usage credits. Anthropic says it will extend the included window “if capacity allows” and aims to make Fable 5 a standard part of subscriptions when it can.
  • Consumption-based Enterprise plans: fully available.

What Anthropic and its customers report

Anthropic’s comparison charts are published as images, so the points below come from the announcement’s text. All are reported by Anthropic or by early-access customers it quotes; none are independently verified.

  • On Cognition’s FrontierCode evaluation, Anthropic says Fable 5 “scores highest among frontier models, even at medium effort.”
  • Customers quoted in the announcement say it achieved the top score on the CursorBench coding benchmark and on Hebbia’s Finance Benchmark.
  • One customer reports it was the first model to exceed 90% on its core analytics benchmark, a 10-point gain over Opus.
  • For Mythos 5, Anthropic reports that it accelerated some aspects of drug design by around 10 times, and that scientists preferred its molecular biology hypotheses about 80% of the time in blinded comparisons with Opus-class models.

Who it’s for

A good fit: teams running long, complex coding or knowledge-work tasks, where the per-token premium may be offset if the model finishes work in fewer attempts. The 1M-token context and 128K output limit suit large codebases and long documents.

Worth weighing carefully:

  • Security and life-sciences work. Requests in these areas will often be answered by Opus 4.8 instead, so organisations whose day-to-day work sits in these domains may not see Fable-level output there. Anthropic’s route for such users is its gated Mythos programmes.
  • High-volume, cost-sensitive workloads. At double the per-token price of Opus 4.8, Fable 5 makes most sense where its extra capability changes the outcome.
  • Strict data-retention policies. The 30-day retention requirement will need checking against internal rules before adoption.
  • Subscription users. The included-access window is short, and Anthropic has tied its extension to capacity.

What it means

Our take: rather than hold back a more capable model, Anthropic has drawn lines around specific risk areas and routed those requests elsewhere. Its own figures suggest most commercial users will rarely notice. Users near those lines will get something closer to Opus 4.8, and the company itself says the biology and chemistry boundary is wider than it would like.

As with any vendor launch, the benchmark claims describe the best case. Buyers should measure the fallback rate and cost per completed task on their own workloads before deciding whether the premium over Opus is justified.

Sources

See All

More on Anthropic

See All

Type to search. ↑ ↓ to navigate, ↵ to open.