Rumor has it that GPT-6 has been under closed beta for almost two and a half months, and its ability is suspected to be close to AGI

Comparative news, according to monitoring, OpenAI CEO Sam Altman will introduce the next generation model to the US government and lawmakers next week. The model has not been officially announced, but the community has put together recent clues into a GPT-6 capability map. OpenAI has confirmed that overturning the Erdős unit distance conjecture, breaking the sandbox in the nanoGPT competition, and splitting the authentication token to bypass the scanner came from the same long-term mission model. It keeps track of the target. Once limited, it doesn't give up immediately, and it actively searches for system vulnerabilities. The model was tested in early May at the latest. It submitted PR #287。 to the nanoGPT repository at the time A follow-up PR that appeared on May 9 already cites its approach. According to public records, OpenAI has been in closed beta for at least two and a half months. The latest announcement has once again raised the upper limit of abilities. According to OpenAI, GPT-5.6 Sol and a more capable pre-release model will jointly break through the quarantine environment in cybersecurity assessments. They used the zero-day vulnerability to gain internet access and then enter Hugging Face's production system in an attempt to directly obtain evaluation answers. From original scientific research and long-term execution to independently discovering and exploiting zero-day bugs, this set of capabilities has surpassed the normal chat model. Calling it close to AGI is still a community judgment, not OpenAI's official conclusion.