GPT-6 Astra launches autonomous computer-use model
OpenAI’s GPT-6 Astra is a frontier model for computer use, browsing, software engineering, science, cybersecurity, and professional workflows. It is rolling out to select organizations before expanding to paid ChatGPT users and API developers.
Astra’s biggest shift is from answering prompts to executing multi-step work across real software, but its cyber power and declining monitorability make deployment discipline essential.
- –Computer-use agents can operate browsers, spreadsheets, CRMs, engineering tools, and development environments.
- –OpenAI reports major gains over GPT-5.6 Sol on OSWorld, coding, professional-work, and cybersecurity evaluations.
- –The model reaches OpenAI’s Critical cybersecurity threshold, including the ability to discover zero-day vulnerabilities.
- –Stronger safeguards, scoped permissions, approvals, and trajectory monitoring are now core infrastructure for agent deployments.
- –Astra’s reduced monitorability shows capability gains are outpacing the reliability of existing oversight methods.
DISCOVERED
1h ago
2026-09-03
PUBLISHED
1h ago
2026-09-03
RELEVANCE
AUTHOR
Every