OpenAI has introduced GPT-6 Astra, its new flagship model for professional work. The model is available now in ChatGPT Work, Codex and the API. OpenAI positions it as state of the art in computer use, browsing, software engineering, cybersecurity and science.

A key claim is that Astra works through the same desktop applications people use, even without an API. Businesses can deploy it inside existing workflows without custom integrations. In a Financial Modeling World Cup test from the 2023 Microsoft Excel World Championship, Astra completed challenges via computer use about four times as fast as the winning human.

Internally, OpenAI's engineering team used Astra to find a memory-allocation bottleneck in a Codex test environment. Switching allocators cut turn latency by 25 times, at roughly 30 percent higher peak memory use. The marketing team cut three hours of footage into a launch video that passed 550,000 views in four days.

On Terminal-Bench 4.0, which tests agents on terminal-based engineering and data tasks, Astra scores 57.9 percent. GPT-5.6 Sol2 reaches 37.3 percent and Claude Fable 5.1 reaches 55.8 percent, at about 9 percent and 63 percent lower estimated API cost per task. Pricing starts at 10 dollars per million input tokens and 50 dollars per million output tokens.

On safety, OpenAI says Astra produced unintended outcomes 89 percent less often than GPT-5.6 Sol on its internal computer use benchmark. New enterprise admin controls restrict access to approved websites and applications, manage uploads and downloads, and enforce confirmation policies for consequential actions. Astra is the first model to hit the Critical cybersecurity threshold under OpenAI's Preparedness Framework. Enterprise plugins for Oracle Analytics, Power BI, Navan and Avalara ship alongside. Zero Data Retention is available for eligible API customers.