GPT-6.2 Gray Testing Begins, Claude Mythos Exposed

Techub News reports that some users have seen GPT-6.2-Astra appear in the OpenAI model list and it is already callable; some users say that 20% or even 100% of their requests are routed to this model, with the grayscale mainly targeting users of the $500 Pro package. The front-end code of Anthropic's official website briefly showed Claude Mythos 5.5, which was removed after several hours; this version is said to include biological research and cybersecurity directions. A repository claiming to leak the Claude Mythos weights appeared on the model hosting platform HuggingFace, which has been downloaded over 5000 times; the authenticity has not yet been confirmed. (Quantum Bit)