OpenAI has released GPT-6 Astra, its most capable model to date, available in ChatGPT Work, Codex, and the API. The update positions Astra as a state-of-the-art solution for complex professional tasks including software engineering, cybersecurity, and science.
- Astra operates directly within everyday applications without requiring APIs, allowing businesses to integrate AI into existing workflows immediately.
- Internal testing showed a 25x reduction in turn latency after resolving memory-allocation bottlenecks using Astra and Codex.
- The model achieves higher cost-efficiency by completing tasks in fewer tokens with fewer retries, priced at $10 per million input tokens and $50 per million output tokens.
- Astra demonstrated 89% fewer unintended outcomes than GPT-5.6 Sol on internal computer use safety benchmarks.
- New enterprise admin controls allow restriction of access to approved websites and applications, alongside automated review for consequential actions.
The release aims to provide immediate value for demanding professional work while offering stronger alignment and control mechanisms for enterprise deployment.