OpenAI launches GPT‑6 Astra, its most advanced AI model, with record scores in cybersecurity, mathematics, and professional automation. The rollout begins for select organizations and will soon reach all premium users and developers.
OpenAI has released GPT‑6 Astra, its latest AI model, which is already making an impact in cybersecurity, scientific research, and professional automation. Astra is designed to handle real-world tasks that demand both technical skill and sound judgment.
In cybersecurity, Astra scored 100% on ExploitBench and can spot previously unknown zero-day vulnerabilities. The model also achieved 99.9% on ARC-AGI-3 and 98% on FrontierMath Tier 4, and has already helped solve longstanding mathematical problems. These results mark a significant jump in both technical performance and practical use.
Astra is the first OpenAI model to reach Critical-level cybersecurity capabilities under the company’s Preparedness Framework, prompting OpenAI to restrict access to its most powerful cyber features.
For professionals, Astra can take over repetitive digital tasks like filling out online forms, updating CRM records, organizing calendars, and conducting online research. In software engineering, Astra writes code that needs fewer revisions before deployment, outperforming earlier models on internal tests. Early-access organizations are already using Astra to generate documents, spreadsheets, and presentations for complex workflows.
Astra is built for speed and accuracy. It handles demanding professional tasks with a level of judgment and transparency that, according to OpenAI, surpasses previous models. Improvements in alignment mean users can trust Astra with sensitive or complex assignments, as the model better understands user intent and communicates its own limits more clearly.
In science, Astra sets new records across math and science evaluations. It can analyze data, generate plots, and run frontend QA checks for websites, making it useful for both researchers and developers. Astra can also install and test software, troubleshoot on-screen issues, and check that digital products work as intended.
According to OpenAI’s official rollout plan, Astra was initially made available to a limited group of organizations, with broader access for ChatGPT Plus, Pro, Business, and Enterprise users, as well as API and AWS integration, scheduled for the following days. The company also highlights Astra’s state-of-the-art results on Terminal-Bench Science 0.1 and HealthBench Pro, underscoring its strength in scientific and professional scenarios.
Access to GPT‑6 Astra starts today for a limited group of organizations. In the coming days, it will become available to all ChatGPT Plus, Pro, Business, and Enterprise users. Developers can integrate Astra through the OpenAI API and AWS, with usage included in current subscription plans and options for extra credits. Standard API pricing is $10 per million input tokens and $50 per million output tokens, with a Fast mode offered at double the speed and price. According to industry reports on OpenAI’s briefings, company president Greg Brockman said safety was the top priority at this level, and the initial release is limited to select clients.
OpenAI says Astra is deployed with strong safeguards to prevent misuse, especially in cybersecurity. The company reports fewer misaligned outcomes compared to earlier frontier models and highlights Astra’s ability to respect task boundaries and explain its actions. OpenAI’s rollout and pricing are meant to balance access with responsible deployment.
Astra’s launch marks a shift in how AI will be used in daily professional life. Its ability to automate complex workflows, secure digital environments, and speed up scientific work is likely to change what businesses and individual users expect from AI. As OpenAI continues to push the limits of alignment and capability, Astra signals a new phase: autonomous, reliable AI is now part of the workplace, and the competition among advanced models is set to intensify.
