OpenAI introduced GPT-4 Astra, a new frontier model now available in ChatGPT Work, Codex, and the API. The announcement positions it as state-of-the-art on computer use, browsing, professional work, software engineering, cybersecurity, and science, aimed at the most demanding professional work.

The key positioning is that Astra removes integration friction. Most AI systems require businesses to prepare data, redesign workflows, and build custom integrations before they can deliver value. Astra runs in ChatGPT Work and Codex, writing code and working through the same applications people use daily, even when those applications lack an API. The result is that businesses can put it to work within existing workflows from day one. Early customers are reportedly using it to optimize GPUs, spot discrepancies in financial statements, and produce more on-brand decks. OpenAI also rolled Astra out internally weeks before launch: its developer and marketing teams used Astra and Codex to turn three hours of multicamera footage into a video that has drawn over 550k views in four days. Engineering used Astra to identify a memory-allocation bottleneck behind slow Codex sessions; switching allocators cut turn latency 25× with roughly 30% higher peak memory use. The model is also described as better at following a company’s voice, templates, and design standards.

The cost story focuses on efficiency. Astra is trained to complete tasks with fewer tokens and fewer retries, which translates to less rework and lower cost per task. OpenAI says it occupies the majority of the cost-efficiency frontier on professional work and coding evaluations, including Terminal Bench 4.0 and the Artificial Analysis Intelligence Index. Pricing starts at $10 per million input tokens and $50 per million output tokens.

On safety, OpenAI calls Astra its most aligned model yet. It was tested on an internal computer use safety benchmark covering scenarios such as exposing confidential information, sharing a dashboard too broadly, and deleting data. Astra produced unintended outcomes 89% less often than GPT-5.6 Sol and 74.7% less often than Claude Fable 5.1. Adding confirmation steps and automated review further helped. New enterprise admin controls let organizations restrict access to approved websites and desktop applications, manage uploads and downloads, and control browsing history, so teams can start with a limited configuration and expand over time. ChatGPT Work and Codex include confirmation policies before consequential actions and automated review to catch unsafe tool calls.

Astra is also the first model to reach the Critical cybersecurity capability threshold under OpenAI‘s Preparedness Framework. OpenAI says it strengthened protections against misuse and unauthorized actions, including training the model to respect safety boundaries, improving resistance to attempts to bypass safeguards, and deploying automated checks to block harmful responses. Alongside Astra, new enterprise plugins for Oracle Analytics, Power BI, Navan, and Avalara are launching in ChatGPT Desktop, powered by browser-use capabilities. Zero Data Retention is available for eligible API customers on supported endpoints, subject to approval.