OpenAI has rolled out GPT-6 Astra, describing it as the most capable model the company has ever released. The launch began with a small group of partner organizations before opening up to ChatGPT's paid tiers and the OpenAI API "in the coming days," following the staged rollout pattern OpenAI has used for its last few major releases.
What makes Astra different
Astra is being positioned less as a chatbot upgrade and more as a tool for genuinely complex, multi-step work — things like reviewing and patching code, operating a computer to complete a task, and producing finished documents, spreadsheets or presentations rather than just draft text. Independent evaluators reported strong results across reasoning and coding benchmarks compared to OpenAI's previous flagship model.
Why the safety angle matters
The detail that stood out most to industry watchers wasn't a benchmark score — it was a safety classification. Astra is the first OpenAI model to cross into what the company calls the "Critical" tier of cybersecurity capability under its own internal safety framework. In practice, that means the model is judged capable of finding previously unknown security flaws and working out ways to exploit them without much human guidance, which is exactly the kind of capability that also makes a model useful for legitimate security defense work. OpenAI says it has added extra safeguards specifically to manage that risk before wider release.
The bigger picture
Astra's release lands in the middle of a broader moment of scrutiny for frontier AI models generally, with independent AI researchers noting that model capabilities remain "jagged" — extremely strong in some tasks, inconsistent in others — even as the headline numbers keep climbing. For everyday users, the practical impact will show up gradually as Astra becomes available across ChatGPT's paid plans and the wider API ecosystem over the next few weeks.
Reporting on this story also appeared at OpenAI, which has more technical detail if you want to go deeper.