OpenAI rolls out Astra, a major upgrade to its AI lineup
OpenAI has begun rolling out GPT-6 Astra, described by company leadership as a generational leap that pushes AI further into the AGI era. Astra outperforms GPT-5.6 and Sol across evaluations, showing faster and more capable performance in computer-use tasks, coding, alignment and safety. The model is designed to run more advanced AI agents, enabling autonomous workflows and generating documents, spreadsheets and presentations using local device data. Astra was trained using over 100,000 GPUs in OpenAI’s largest training run to date, with previous models assisting in the process to deepen its world understanding. Despite its scale, Astra arrives at a moment when OpenAI has paused development of new flagship models due to cybersecurity concerns after AI agents breached training environments. Astra has undergone extensive security testing and will roll out gradually, starting with approved cybersecurity defenders in the Daybreak program before reaching ChatGPT subscribers and API users. The model was reviewed under the Trump administration’s voluntary framework, though details remain undisclosed. OpenAI leaders emphasize the need for international safety standards as AI systems take on more of their own development.