OpenAI Unleashes GPT-6 Astra: New AI Model Can Take Control of Your Computer

OpenAI has unveiled GPT-6 Astra, its latest artificial intelligence model, introducing a major expansion in what AI systems can do by enabling the model to operate computers and complete complex digital tasks on behalf of users.

The model was announced on September 3, 2026, with OpenAI describing Astra as its most intelligent and aligned model to date. The company said the system represents a new generation of AI capable of handling computer use, browsing, software engineering, cybersecurity, science and professional work.

Unlike conventional chatbots that primarily respond to prompts, Astra is designed to interact directly with computer environments. It can navigate websites, work through software interfaces, manipulate spreadsheets, fill forms and carry out multi-step workflows with limited human intervention.

OpenAI said the model achieved state-of-the-art results on several evaluations, including AutomationBench and ScreenSpot Pro. It also reported scores of 98 per cent on FrontierMath Tier 4 and 99.9 per cent on ARC-AGI-3, highlighting Astra’s performance in demanding reasoning and computer-use tasks.

The model’s capabilities extend beyond routine office tasks. OpenAI said Astra can assist with software engineering, conduct research, work with professional applications and identify serious cybersecurity vulnerabilities. The company has consequently classified the model as reaching a critical threshold in cybersecurity capability.

OpenAI is also presenting Astra as a more carefully controlled system. The company said improvements in alignment have made the model better at understanding user intent, respecting task boundaries and avoiding actions beyond what it has been authorised to do.

The release nevertheless comes amid increasing debate over the safety of highly autonomous AI agents. The ability to interact directly with computers creates new possibilities for productivity but also raises questions about monitoring, cybersecurity and how much control should be delegated to an AI system.

OpenAI said GPT-6 Astra was initially being rolled out to a limited group of organisations, with access for ChatGPT Plus, Pro, Business and Enterprise users expected in the following days. The model is also being made available through the OpenAI API and cloud platforms including Microsoft Azure and AWS Bedrock.

The launch strengthens the growing shift from AI that simply generates information to AI agents capable of taking action. Instead of giving users instructions on how to complete a digital task, systems such as Astra are increasingly being designed to perform the task themselves from beginning to end.

With GPT-6 Astra, OpenAI is therefore pushing computer-operating AI further into mainstream use, potentially changing how people interact with websites, software and digital workplaces as increasingly sophisticated AI systems move from answering questions to carrying out real-world computer tasks.