OpenAI has launched GPT-6 Astra, a new AI model designed to improve reasoning, computer use, and autonomous task execution. According to the company, Astra sets new performance highs across mathematics, scientific research, software engineering, browser automation, and cybersecurity, while outperforming GPT-5.6 Sol on several complex agent-based tasks.
The model can work through longer workflows instead of handling only individual requests. It can research information online, operate software, analyze scientific data, create websites and applications, troubleshoot technical issues, and prepare business documents, spreadsheets, and presentations. Astra is also better at handling ambiguous instructions and staying focused as tasks evolve.
Cybersecurity is among the model's biggest advances. OpenAI reports substantially stronger performance in vulnerability analysis and exploit-related testing, but says these capabilities require stricter safety measures. The production version is intended to support defensive cybersecurity work while restricting potentially harmful advanced requests.
GPT-6 Astra is starting with a limited rollout and will later expand to ChatGPT Plus, Pro, Business, and Enterprise users, as well as the OpenAI API, Microsoft Azure, and AWS Bedrock. Pro, Business, and Enterprise customers will also receive access to GPT-6 Astra Pro.