Anthropic and the Oceanus Leak: A New Model That Intrigues

A new artificial intelligence model, potentially named Oceanus, has recently emerged at Anthropic, generating considerable interest. This unexpected discovery comes as the Mythos model continues to spark debate within specialized circles.

On June 3, observers spotted an unusual identifier, claude-oceanus-v1-p, in Anthropic's Claude environment. This finding quickly circulated on social media and forums, fueling speculation about a successor to Mythos, a model already known for its sensitivity and restricted access.

Researchers have also reported finding references to Oceanus through unofficial proxy services, which has bolstered the idea of a leak. Among them, Lisan al Gaib mentioned on Twitter that they discovered an API offering Oceanus, with pricing that appears more reasonable than that of Mythos, estimated at $16 for input and $80 for output per million tokens. Oceanus is said to display a throughput of about 52 tokens per second, placing it in the high-end model category.

The situation is even more delicate for Anthropic as unauthorized access to the model is said to have been resold shortly after testing began. This leak comes as the company is conducting a red teaming phase to assess the risks associated with Oceanus.

Although no official announcement has been made by Anthropic regarding Oceanus, the model seems directly linked to Mythos Preview, presented in April 2026 as part of the Glasswing program. This information, stemming from leaks, continues to captivate the attention of cybersecurity experts and enthusiasts of advanced technologies.