GPT‑6 Astra cuts unintended outcomes 89% vs GPT‑5.6 Sol

OpenAI positions Astra as a model that fits into existing workflows rather than requiring businesses to rebuild them. The company says Astra can write code and work through the same applications people use daily, even when those applications lack an API. It also highlights early customer uses, from optimizing GPUs to spotting discrepancies in financial statements, and internal tests such as turning three hours of multicamera footage into a video that drew over 550k views in four days.

The launch pairs capability gains with control features. Astra is the first model to reach the Critical cybersecurity capability threshold under OpenAI's Preparedness Framework, and OpenAI says it strengthened protections against misuse and unauthorized actions. At the same time, new enterprise admin controls let organizations limit access to approved websites and apps, manage uploads and downloads, and require approval before consequential actions.

For teams weighing adoption, the outcome may hinge on whether those safeguards and controls work as intended when Astra is given access to consequential business systems. The stated safety results and the Critical designation point in opposite directions, so the practical test is whether organizations can start with a limited configuration and expand access without unexpected incidents.