Anthropic released Claude Mythos publicly as 'Fable' at 2x Opus pricing, targeting agent workflows with strong safety limits.
Anthropic released Claude Mythos publicly today as 'Fable' at 2x Opus pricing. The model, previously restricted to Project Glasswing partners since April 2026, now opens to general availability with safety limits.
Key facts
- Public Mythos version codenamed 'Fable' released today
- Priced at 2x Opus, down from initial 5x preview cost
- Strong safety limits; no cyber capabilities for public
- Project Glasswing partners found thousands of vulnerabilities
- Mythos preview launched April 2026
Anthropic is dropping a public version of Mythos today, codenamed 'Fable,' per The Information. The model arrives at 2x the price of Opus — still expensive, but a significant discount from the initial Mythos preview pricing of 5x Opus that sparked sticker shock among enterprise buyers in April.
Restricted safety, no cyber access
Fable comes with strong safety limits and will not be as open on cyber use as the restricted preview given to Project Glasswing partners [According to @rohanpaul_ai]. Those partners reportedly found thousands of major vulnerabilities during the preview period, including zero-day exploits. Anthropic is deliberately narrowing the public release's capabilities in cybersecurity to reduce misuse risk.
Agent workflows as the differentiator
The model is expected to be much stronger at long-running, multi-step tasks and agent-style workflows [per the same source]. This positions Fable directly against OpenAI's o-series reasoning models and Google's Gemini 2.0 agent capabilities, which have been vying for enterprise agent contracts throughout 2026. The 2x Opus price tag suggests Anthropic is betting that superior task completion reliability will justify the premium, rather than raw benchmark scores.
Context on Mythos
Anthropic introduced Claude Mythos Preview in April 2026 as its most powerful frontier model, especially strong in coding, reasoning, and cybersecurity — including finding and exploiting zero-days [As previously reported]. It was not released publicly at first because of safety issues. Only selected Project Glasswing partners received access for defensive cybersecurity, and they have reportedly found thousands of major vulnerabilities. The public version now arrives roughly three months after that preview, with a narrowed safety envelope and a pricing structure that suggests Anthropic is still figuring out how to monetize frontier capability without triggering regulatory backlash.
What to watch
Watch for enterprise adoption rates in agent workflows over the next 60 days, and whether Anthropic adjusts safety limits based on real-world misuse reports. Also track any pricing changes to Opus or Sonnet as Anthropic repositions its model tiering around Fable.
[Updated 10 Jun via the_verge_tech]
On SWE-Bench Pro, which penalizes incorrect coding agents, Claude Fable 5 scored 80.0%, compared to Claude Opus 4.8's lower score [per Towards AI]. In a real-world test, Fable 5 completed a Stripe code migration in one day that would have taken a team two months [per The Decoder]. Anthropic also released a separate Mythos 5 model, which designed drug candidates autonomously but remains locked down due to offensive cyber capabilities [per The Decoder]. Claude subscribers can try Fable 5 until June 22 without spending usage credits [per Engadget].
Originally published on gentic.news
Top comments (0)