Anthropic's new models are coming

Anthropic just accidentally leaked their own roadmap. A basic CMS misconfiguration left nearly 3,000 unpublished internal assets completely public. Security researchers (and the leaker m1astra) walked right in and found the blueprints for Anthropic’s next major leap.

Here is the fast, no-fluff breakdown of what we actually learned.

[The Core Facts]

• The New Tier: This is not an incremental update to Opus. The leaked draft blogs reveal a brand-new, massive model tier sitting above Opus. Internally, the drafts refer to it interchangeably as Capybara and Mythos.

• Next-Level Cyber Skills: The model reportedly crushes Opus 4.6 in reasoning and coding. More importantly, Anthropic internally warns it is "far ahead of any other AI model in cyber capabilities." The revelation was significant enough that major cybersecurity stocks dropped following the leak.

• The Defender Rollout: Because this model could easily be weaponized to discover and exploit zero-day vulnerabilities, Anthropic is locking it down. They are running a slow Early Access Program exclusively for "cyber defenders" so security teams can patch vulnerabilities before the public gets access.

• The Price Tag: It is extremely compute-heavy. Anthropic openly admits in the drafts that the model is "very expensive for us to serve, and will be very expensive for our customers."

   

 

Check the Research Preview