Anthropic released Claude Fable 5 on Tuesday, the first broadly available model from its Mythos class — a tier the lab had previously deemed too dangerous to ship because of how well it found cybersecurity weaknesses. Pricing lands at $10 per million input tokens and $50 per million output tokens, twice the cost of Claude Opus 4.8 and half what users already pay for the gated Mythos Preview. The release comes roughly a week after Anthropic confidentially filed IPO paperwork.
The unlock, according to Anthropic, is a new safeguard layer that blocks responses in specific high-risk areas — chiefly cybersecurity and biology — and falls back to Opus 4.8 when a prompt trips a classifier. The company said early data shows 95% of Fable sessions run entirely on Fable's own responses, with only the remaining 5% routed to the smaller model. Anthropic said it red-teamed those classifiers against jailbreaks before shipping.
Dianne Penn, Anthropic's head of product management, research and labs, framed the timing as a confidence call rather than a capability one. Mythos itself was first revealed in March through a data leak and formally announced in April, when Anthropic called it a "step change" and restricted access through Project Glasswing, an initiative aimed at letting cybersecurity professionals harden defenses against AI-driven threats.
“The reason why we're releasing Fable Five now is very much due to us feeling more confident with our safety guardrails in place”— Dianne Penn, Anthropic's head of product management, research and labs
Key facts
- 01Anthropic released Claude Fable 5 on June 9, 2026, the first public model from its Mythos class.
- 02Pricing is set at $10 per million input tokens and $50 per million output tokens, double Claude Opus 4.8 and half what Mythos Preview costs.
- 03Anthropic said 95% of Fable sessions ran entirely on Fable's own responses, with the rest routed to Opus 4.8.
- 04High-risk prompts in cybersecurity and biology fall back to Opus 4.8 under new safeguards.
- 05Claude Mythos 5 ships only to vetted partners already inside Project Glasswing.
Fable 5's pitch is long-horizon work. Penn said the model is materially better at "long-horizon memory management" — keeping track of multi-step tasks where prior Claude models would lose the thread partway through. She cited improvements in self-verification and instruction following, with the model checking its own work and course-correcting before returning a final answer.
Anthropic said Fable 5 "shows exceptional performance in software engineering, knowledge work, and vision," and that its lead over competing models widens as task length and complexity grow. Penn pointed to overnight autonomous projects — running code-base reviews end-to-end — as the kind of work the lab is now telling customers to push at the model.
Alongside Fable 5, Anthropic is shipping Claude Mythos 5 to vetted partners who already have access to Mythos Preview. Per Anthropic's blog, Mythos 5 is the same underlying model as Fable 5 "but with the safeguards lifted in some areas." The lab said Mythos Preview users will be able to upgrade and that it plans to expand access through what it called a more systematic trusted-access program.
“We recommend that customers give their most challenging work to Fable Five”— Dianne Penn, Anthropic's head of product management, research and labs
The pricing structure is worth sitting with. At $50 per million output tokens, Fable 5 is the most expensive Claude that any developer can call without a partnership review — twice Opus 4.8 — but it is also priced as a discount to the Mythos Preview tier those partners are currently paying. Anthropic is effectively setting a public ceiling below its private one, betting that the 95% in-class hit rate justifies the markup for long-running agentic work.
Anthropic has faced criticism in the past for safeguards that over-block benign prompts, and the company acknowledged that pattern is likely to repeat in the first weeks of Fable 5's rollout. Penn said safeguard improvements will land post-launch. The naming also raised questions: both models ship as "5" without any prior public Mythos or Fable releases, and Anthropic declined to explain on the record how the two relate to Mythos Preview.
“We recognize that there might be some benign requests that end up being blocked initially”— Dianne Penn, Anthropic's head of product management, research and labs
The release reframes Anthropic's safety posture from "hold it back" to "ship it with a fallback router," and it sets a template the lab can reuse on future capability jumps. If Fable 5 holds the line on cybersecurity and bio prompts in the wild, Anthropic gets to commercialize a model class it was previously sitting on — useful revenue heading into an IPO process — while keeping a more permissive Mythos tier reserved for partners willing to be vetted. The pricing, the routing, and the dual-SKU structure are all moves competitors will study.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
