OpenAI launches GPT-6 Astra

OpenAI announced GPT-6 Astra on September 3, presenting a general-purpose model built to work across computer interfaces and longer workflows. Its scope extends from answering requests to carrying out connected tasks in browsers, software tools, and professional applications.

From online forms to software testing

OpenAI says Astra can fill online forms, update CRM records, organize calendars, research online, and draft summaries inside email or document editors. It can also analyze scientific data, generate plots, build websites, run frontend QA, install and test software, and troubleshoot problems shown on screen.

Access is expanding in stages

The initial rollout began with a limited group of organizations on September 3. OpenAI said access would expand over the following days to ChatGPT Plus, Pro, Business, and Enterprise users, as well as the OpenAI API and AWS, so availability may differ between eligible accounts. GitHub separately announced general availability in Copilot on September 4 with a gradual rollout for Pro+, Max, Business, and Enterprise users across its model picker in supported editors, apps, agents, command-line tools, and GitHub services. Business and Enterprise administrators can manage access through model policy settings.

Critical capability brings stricter safeguards

OpenAI classifies Astra as its first model to reach the Critical level for cybersecurity capability under its Preparedness Framework. The company says that with suitable tools and access, the model can discover previously unknown flaws and develop new exploitation methods across many well-protected systems without step-by-step human guidance. Its deployment protections include stricter isolation, checkpoint encryption, full-trajectory monitoring that includes chains of thought, and a blocking alignment evaluation process before internal use. OpenAI also cautions that monitorability has decreased compared with GPT-5.6 Sol because Astra has greater control over its chain of thought and is less likely to include incriminating information there. This is a monitoring limitation identified by OpenAI, not evidence of harmful intent.