Anthropic Releases Fable 5.2 Widely and Mythos 5.2 by Approval
Anthropic has released two Claude models at once and split them by who is allowed to use them, selling Fable 5.2 to anyone while keeping Mythos 5.2 behind a verification wall for cybersecurity and life sciences work.
Fable 5.2 is the general model and targets coding, reasoning and agent tasks, which is where the commercial fight is. Mythos 5.2 is aimed at approved cybersecurity and life sciences research, and Anthropic says access is restricted to organisations that pass its verification programmes rather than to whoever pays.
And the pricing moved in the buyers' direction. Both models cost USD 8 per million input tokens and USD 40 per million output tokens, which is cheaper than the previous Fable and Mythos releases, so the company is competing on price at the same moment it is competing on capability.
The performance claim is narrower than a benchmark sweep. Anthropic says the two models generate output up to 30 per cent faster than their Claude 5.1 predecessors and communicate more clearly with users, which are the numbers that matter to anyone running them inside a product rather than testing them in a demo. Neither is described as a new architecture, so the release is framed around speed and clarity rather than a jump in raw capability.
The launch lands before a planned IPO, and it follows Claude Haiku 5.5, so this is the second model release in two months from a company assembling a full product line ahead of a listing rather than shipping one flagship a year.
But the choice to gate Mythos is the part worth watching. A model that writes exploit code or designs biological experiments carries a different liability from one that writes software, and a verification programme is Anthropic's answer to selling capability in those fields without selling it to everyone.
Rivals are moving in the same direction on capability. OpenAI, Google and Meta have all pushed models tuned for coding, reasoning and agents to win developers and enterprise accounts, which is why a launch this routine still lands in a crowded month. Meta and Google have both pushed agent tooling at developers recently, so access and price carry as much weight here as the model card.
So read the split as product strategy rather than ethics alone. A single model family sold two ways lets Anthropic charge enterprise prices for vetted access while keeping a cheap general model in front of developers, and the missing piece is how many organisations actually clear verification.
Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 together
Fable 5.2 covers coding, reasoning and agent tasks and is broadly available
Mythos 5.2 is limited to organisations approved for cybersecurity and life sciences work
Anthropic says output is up to 30 per cent faster than the Claude 5.1 models
Pricing is USD 8 per million input tokens and USD 40 per million output tokens
Source: Reuters
