Anthropic has released a version of its Claude Mythos AI model to the general public, the company confirmed this week. The rollout follows an extended period of controlled testing and marks the first time everyday users can access what Anthropic has described as its most capable and safety-evaluated model class to date. The Guardian first reported the release, noting the company's particular emphasis on safety procedures as a prerequisite for going public.
What the Release Means for Users
The public launch arrives after a period of significant anticipation. Anthropic had teased the Mythos model earlier this year while the company was riding a wave of investor confidence, with its valuation climbing toward the trillion-dollar range. Since then, the company has been methodical about widening access, prioritizing safety audits and partner feedback before opening the floodgates. Users can now access the model through Claude.ai and the Anthropic API, though some capability tiers remain tied to subscription plans.
Key Facts
- Claude Mythos is now available to the general public via Claude.ai
- Anthropic conducted extensive safety evaluations before the public launch
- The model was previously accessible to select partners and researchers
- API access is available for developers building on the Mythos model class
- The release follows the earlier Claude Fable launch under the Claude 5 generation
Anthropic has been transparent about the safety work underpinning this release. The company ran the model through its standard battery of evaluations, covering areas like biosecurity, cybersecurity, and what Anthropic calls "autonomy" risks, where a model might take unsanctioned actions if deployed in agentic settings. The results, the company says, cleared its internal thresholds for safe public deployment. That framing has become central to how Anthropic positions Mythos, distinguishing it from competitors who have faced criticism for moving faster with less visible safety documentation.
We believe frontier AI must be developed with safety as a prerequisite, not an afterthought. Mythos represents our best effort to hold both capability and responsibility together.Anthropic, company statement
A Staged Road to Public Access
The path to today's launch was deliberately paced. Project Glasswing allowed select partners to share Mythos findings ahead of the broader release, giving Anthropic real-world signal about how the model performed outside controlled lab conditions. That data informed tweaks to the model's behavior before it was handed over to the general public. The earlier release of Claude Fable as the first Claude 5 model also helped set expectations about what the Mythos tier would offer and how it fits within the broader lineup.
For developers, the Mythos release opens up a new tier of capability for building applications. Anthropic has positioned the model as suited for complex reasoning tasks, long-context work, and agentic use cases where a model must plan and execute multi-step tasks. Pricing details for API access have been published on Anthropic's developer platform, with costs reflecting the model's position at the top of Claude's model family.
What Comes Next
The public launch does not necessarily mean the Mythos story is complete. Anthropic has signaled that model updates and capability expansions are part of the plan going forward, and the company continues to invest in interpretability research intended to give engineers a clearer picture of what is happening inside these systems. How users and developers actually put Mythos to work over the coming weeks will shape what adjustments Anthropic makes in subsequent updates.
For now, the release represents a concrete step in Anthropic's broader strategy: build capable models, validate them rigorously, and expand access incrementally. Whether that approach continues to satisfy users hungry for the fastest and most capable AI tools, while also reassuring policymakers and researchers worried about safety, remains an ongoing test. The public launch of Mythos is, in that sense, both an endpoint and a beginning.
