OpenAI launches GPT-6 Astra, its most powerful model yet and touts its ability to use your computer

OpenAI has unveiled GPT-6 Astra, its most advanced model to date, boasting powerful new capabilities. One standout feature is "computer use," allowing the AI to seamlessly interact with computers as humans would. This includes navigating spreadsheets, filling out forms, and browsing webpages at superhuman speed. Co-founder Greg Brockman highlighted this as a significant advancement, stating the model can perform tasks "at superhuman speed."

OpenAI showcased Astra's prowess with benchmark tests, outperforming previous models such as GPT-5.6 Sol and Anthropic's Claude Fable 5.1. Notably, Astra achieved a remarkable 98.6% score on ARC-AGI-3, a benchmark measuring the model's ability to reason in novel situations, compared to 7.8% for GPT-5.6 Sol and 30% for Claude Opus 5.

Astra also scored 100% on ExploitGym, a cybersecurity challenge, whereas GPT-5.6 Sol scored 78.5%. While Anthropic pioneered AI agents capable of using computers, OpenAI's entry is still not mainstream, but could mark the beginning of the "AGI era," according to Brockman. However, the release of Astra is currently limited to select enterprise customers in OpenAI's cybersecurity-focused Daybreak program.

OpenAI has added rigorous cybersecurity safeguards to the model, including additional monitoring and isolation of training environments, following the Hugging Face incident in July. It's the company's first model to meet its internal "critical cybersecurity capability threshold," allowing it to discover and exploit security vulnerabilities autonomously, under certain conditions.