OpenAI has introduced GPT-6 Astra, a new model designed to advance intelligence in computer use, software engineering, and cybersecurity. The model is rolling out to ChatGPT Plus, Pro, Business, and Enterprise users, as well as via the OpenAI API, Microsoft Azure, and AWS Bedrock.
- GPT-6 Astra saturates FrontierMath Tier 4 with a 98% score and ARC-AGI-3 with a 99.9% score.
- It achieves state-of-the-art computer use performance in OSWorld 2.0, scoring 72.6% in roughly 40 minutes per task, which is about 47% less time than GPT-5.6 Sol.
- The model improves coding efficiency by preserving context across windows in Codex without repeated compaction.
- In cybersecurity evaluations, Astra scored 100% on ExploitBench and discovered two previously unknown zero-day vulnerabilities during testing.
- Alignment improvements include a 0% rate of going beyond authorized targets in internal tests compared to 48% for GPT-5.6 Sol without safeguards.
The release aims to provide users with faster, more accurate tools for complex professional workflows while maintaining strict safety boundaries and alignment with user intent.