Casablanca – OpenAI has introduced GPT-6 Astra, a new model designed for coding, computer use, scientific research, cybersecurity, and complex professional work. The company describes Astra as its most intelligent and aligned model to date.
The model is being rolled out first to a limited set of organizations. OpenAI said it will become available over the coming days to ChatGPT Plus, Pro, Business and Enterprise users, as well as through its API, Microsoft Azure and Amazon Bedrock.
Astra scored 98% on FrontierMath Tier 4 and 99.9% on ARC-AGI-3, according to OpenAI. The company also said the model achieved a 100% score on ExploitBench, which tests whether AI systems can turn known software vulnerabilities into working exploits.
Computer use is another major focus. Astra can fill out online forms, update customer records, organize calendars, conduct online research, create websites, and run frontend quality checks. It can also install and test software and troubleshoot problems shown on screen.
In OpenAI’s OSWorld 2.0 testing, Astra scored 72.6%, compared with 65.7% for GPT-5.6 Sol. OpenAI said it achieved that performance in roughly 40 minutes per task, compared with about 75 minutes for Sol.
The model also received training for professional workflows. OpenAI said Astra can produce documents, spreadsheets and presentations that follow existing templates and instructions. It can adjust when requirements change and ask questions when missing information could affect the outcome.
Read also: OpenAI Pauses Frontier AI Training Over Astra’s Critical Cyber Capabilities
For software developers, OpenAI called Astra its strongest model for software engineering so far. A new feature in Codex will also allow Astra to preserve notes and retrieve information from earlier context windows instead of repeatedly compressing long sessions into summaries.
Astra’s cybersecurity capabilities have prompted additional safeguards. OpenAI said the model meets the Critical threshold for cybersecurity under its Preparedness Framework. In testing without production safeguards, Astra discovered and used two previously unknown zero-day vulnerabilities, which OpenAI said it is disclosing to the relevant maintainers.
The company said Astra will refuse some advanced cybersecurity requests, including creating proof-of-concept exploits for vulnerabilities. OpenAI plans to expand access to defensive cybersecurity workflows in the coming weeks.
OpenAI also reported stronger alignment and safety performance compared with GPT-5.6 Sol. In one evaluation involving more than 54,000 internal Codex tasks, Astra received roughly half as many higher-severity misalignment flags as Sol.
For API users, standard pricing is $10 per million input tokens and $50 per million output tokens. Fast mode is available at twice the standard price and can provide up to twice the speed.
