OpenAI’s GPT-6 Astra model began rolling out on September 3 to paid ChatGPT subscribers following a limited testing phase. The model completes tasks 47% faster than its predecessor GPT-5.6 Sol and scored a perfect 100% on the ExploitBench cybersecurity benchmark, making it the first model to reach the « Critical » threshold in OpenAI’s Preparedness Framework. Its ability to autonomously detect and exploit unknown vulnerabilities led OpenAI to restrict access to the most advanced cybersecurity features during this initial phase. CEO Sam Altman apologized on September 4 for access issues encountered by subscribers. Astra is included in existing subscription tiers at no additional cost.
Source: Read the original article
