OpenAI Launches GPT-6 Astra: 1.05M-Context Computer-Use Model With Safety Gating

OpenAI has released GPT-6 Astra, its most advanced model to date, featuring a 1.05 million token context window and native computer-use capabilities that allow the model to autonomously interact with desktop and web interfaces. Access to the model's most capable tiers is gated behind what OpenAI describes as a 'Critical' cyber threshold — a safety evaluation that determines how much autonomous access a user or organization can unlock. The model represents OpenAI's formal entry into what the company calls the 'AGI era,' signaling a qualitative shift in claimed capability over prior GPT generations. For developers, Astra's long context and computer-use features open new possibilities for document processing pipelines, automated QA workflows, and agentic task execution with far less scaffolding. The safety gating mechanism is also a significant architectural precedent — developers building enterprise integrations should plan for capability tiers based on compliance posture.