OpenAI Launches GPT-6 Astra, Its Most Powerful AI Model
OpenAI announced the launch of its new model, GPT-6 Astra, in a move the company described as a transition to a new generation of artificial intelligence capable of handling complex, multi-step tasks, ranging from research, reasoning, and programming to computer use and completing professional workflows from the initial request to the final result. The company stated that the new model is the most capable among its widely available offerings, able to execute tasks such as web search, filling out electronic forms, updating customer records, scheduling appointments, analyzing scientific data, building websites, testing software, and identifying issues displayed on screen. It can also prepare documents, spreadsheets, and presentations according to specific templates and instructions.
According to tests published by OpenAI, GPT-6 Astra demonstrated faster computer usage, completing simulated tasks in approximately 47% less time compared to GPT-5.6 Sol, while achieving higher scores in task execution. The company also confirmed that additional updates increased the speed of completing certain tasks by roughly 1.9 times compared to the previous iteration. The model features a context window of 1.05 million tokens and can generate up to 128,000 tokens, enabling it to handle large volumes of information and complex projects. It targets areas such as reasoning, programming, research, computer use, and document creation.
However, these new capabilities have raised unprecedented security concerns. OpenAI announced that Astra is the first of its models to reach the “Critical Capabilities” level in cybersecurity according to the company’s preparedness framework. This designation necessitated the implementation of stronger safeguards and protective measures during development and prior to launch, due to the model’s advanced ability to discover unknown security vulnerabilities and develop methods to exploit them when appropriate tools and permissions are available.
GPT-6 Astra is being rolled out in a limited capacity to select institutions, with broader availability to be gradually expanded over the coming days. OpenAI views the new model as a significant step toward AI systems that do more than simply answer questions, but can execute entire chains of actions and take multiple steps to accomplish tasks.
