Anthropic built a new model it describes as a step change in AI capability. Most users won't be allowed to try it.
The model is called Mythos Preview. According to Anthropic, it won't be available to the general public - which makes it unusual among recent releases. Details on who gets access and under what conditions haven't been fully published.
Why Labs Gate Frontier Models
"Step change" is deliberate language in AI announcements. It signals something qualitatively different from existing models, not just a speed or accuracy improvement. For Anthropic, which tends toward careful framing in public communications, that choice of words carries weight.
Restricted access at this level typically comes from two places: cost and caution. More capable models are significantly more expensive to run per query than standard API access - the compute required can be many times higher. Models with substantially expanded capabilities in reasoning or multimodal tasks (handling images, documents, and audio alongside text) may also require more careful deployment conditions than a general-purpose API endpoint.
Access Trajectory to Watch
Anthropic's previous model rollouts followed a consistent path: limited enterprise or research preview, then broader paid API access, then integration into consumer-facing plans. The question with Mythos is whether that rollout is even planned, or whether this model stays permanently restricted.
Claude API customers should watch enterprise tier announcements in the coming weeks. The gap between what Anthropic can currently build and what's available in a standard claude.ai subscription just widened - though by how much depends on details that haven't been published yet.