OpenAI unveils GPT‑6 Astra: new flagship model with a focus on safety
OpenAI has officially unveiled GPT‑6 Astra, its new flagship model. The launch announcement indicates phased access: initially a limited group of organizations will get it, followed by all ChatGPT Plus, Pro, Business, and Enterprise subscribers, as well as API, Azure, and Bedrock users.
According to the company, Astra has improved in computer and browser control, code writing, cybersecurity, and scientific tasks. These improvements are presented as manufacturer claims based on internal tests and do not guarantee similar results under all conditions.
Special attention is paid to security and permission compliance. OpenAI claims that during evaluation for going beyond the assigned task, Astra never exceeded permissible permissions, whereas the previous model GPT-5.6 Sol did so in 48% of cases. This refers to the results of a specific test, not absolute protection.
The model is also oriented toward office tasks, including working with CRM forms, conducting research, and preparing documents. At the same time, OpenAI reminds that even with the model’s high reliability, external precautions remain important.
A phased launch will allow the company to first test the model in real scenarios at partner organizations, and then expand access for mass users and developers. This is standard OpenAI practice for flagship releases.
GPT-6 Astra continues the GPT model line and, according to OpenAI, represents a step forward in reliable instruction following for complex tasks in digital environments.
Primary source: openai.com ↗