OpenAI has deployed GPT-6.2 Astra, the first externally deployed model to reach the High self-improvement threshold in the company’s Preparedness Framework. Released on November 17, 2026, the model is more capable than GPT-6 Astra and scores 5.2 points higher on Agents’ Last Exam, while costing less at $8 per million input tokens and $40 per million output tokens for general users. Its sibling model, GPT-6.2 Sol, is a cost-effective option for everyday tasks that reaches the Critical threshold for cybersecurity and the High threshold for biological and chemical risks. Both models are available from launch in ChatGPT Work, Codex and the API. However, these classifications come from OpenAI’s own framework and do not constitute an external audit.
Source: Read the original article


