OpenAI released GPT-6 Astra on September 3, 2026, calling it the most intelligent and capable model the company has ever built. President Greg Brockman closed the launch briefing with a bold claim: “Welcome to the AGI era.”
Astra is built around three big focus areas: autonomous computer use, software engineering, and long, multi-step agent work. In practice, that means you can ask it to handle a full task — like researching a topic, building a spreadsheet, and turning it into a presentation — and it will plan the steps, operate different tools, and carry the work through to the end largely on its own. On the OSWorld 2.0 benchmark, which tests how well an AI can operate a real computer, Astra scored 72.6%, completing tasks roughly 47% faster than OpenAI’s previous model.
The rollout is happening in stages. Companies in OpenAI’s “Daybreak” cybersecurity program got access first, since Astra is also rated at a “Critical” cyber capability level — meaning OpenAI has built in extra safeguards to stop the model from being used for serious security threats. From there, access is expanding to ChatGPT Pro, Enterprise, and Business Premium users inside ChatGPT Work and Codex, with Plus subscribers following within the ChatGPT Plus subscription at no extra cost. The model is also live via the OpenAI API, Amazon Bedrock, and Microsoft Azure.
Pricing for developers is set at $10 per million input tokens and $50 per million output tokens, with a much larger context window than previous models — up to 1.05 million tokens.
Free ChatGPT users don’t have access yet, and OpenAI hasn’t said when — or if — that will change.