Overview
- OpenAI publicly launched GPT-6 Astra this Sunday and began a staged rollout that puts the model live in the API, Codex and ChatGPT Work for Pro, Enterprise and Business Premium customers while Plus subscribers receive it gradually.
- Astra is built to operate end-to-end on real computers, performing tasks such as filling forms, updating CRM records, running frontend QA, troubleshooting by watching the screen, and keeping state across long coding sessions.
- OpenAI says internal testing pushed Astra to its highest 'Critical' cybersecurity tier because the model can find and chain previously unknown vulnerabilities, so the company is gating exploit-capable features to vetted partners and adding stronger sandboxes and monitoring.
- Benchmark claims in OpenAI’s launch materials show Astra leading on many computer-use and coding tests, but the numbers are self-reported and Astra does not outperform Anthropic’s Claude on every measure, so independent verification is still needed.
- OpenAI set premium API rates for Astra that signal a focus on high-value, stateful workflows and the launch has provoked controversy among creative communities over advertising and IP use, which could reshape how designers and studios adopt agentic tools.