OpenAI has taken another leap forward in artificial intelligence with the introduction of GPT-6 Astra — a model that marks a new stage in human-computer interaction. This advanced development has an unprecedented ability to handle complex tasks autonomously, integrating directly into workflows on both the computer and the web browser.

“We aim to provide tools that allow users to focus on the creative and strategic aspects of their work, delegating routine and technical operations to artificial intelligence,” OpenAI representatives said. This means that tasks such as quickly filling out CRM systems, thoroughly testing software for bugs, and identifying potential vulnerabilities in code can now be automated.

Revolutionary Astra capabilities: from forms to cybersecurity

GPT-6 Astra demonstrates an impressive range of capabilities that dramatically change perceptions of AI’s potential in everyday work. Among the key features that set Astra apart is its deep integration with the computer environment and web browser. The model is capable of:

• Automating work with online forms: from submitting applications to filling out questionnaires, Astra will take over the monotonous work.
• Efficiently updating databases: it will work flawlessly with CRM systems, ensuring information stays up to date.
• Integrating with calendars: scheduling meetings, reminders, and event synchronization will become much simpler.
• Conducting comprehensive online research: Astra can collect, filter, and summarize information from the web.
• Performing deep data analysis: the model can process large datasets, identifying key trends and patterns.
• Creating and testing digital content: Astra can generate websites, create charts, test their functionality, and even install the required software.

To confirm the effectiveness of the new model, OpenAI cites results from testing in the OSWorld 2.0 simulation. There, GPT-6 Astra demonstrated a striking advantage over the previous version, GPT-5.6 Sol. Tasks were completed, on average, 47% faster. Specifically, Astra achieved a 72.6% task success rate, spending about 40 minutes on each task. By contrast, GPT-5.6 Sol posted a 65.7% result with an average completion time of about 75 minutes.

For software developers, OpenAI has introduced an important innovation — a context retention feature. This means that during long coding sessions, the system automatically remembers key details and takes notes, preventing the loss of important information and making the development process easier. This feature will also be available in the updated version of Codex, which has received a new mechanism for working with long sessions.

The evolution of Astra’s training and safety

Training for GPT-6 Astra was aimed at deepening its expertise in working with professional documents, spreadsheets, presentations, and complex multi-step processes. According to OpenAI, the model showed significant improvement in its ability to follow given templates, structure complex documents and presentations. In addition, Astra learned to use only the context relevant to a specific task more effectively, avoiding unnecessary information.

Another breakthrough is the model’s improved handling of ambiguous instructions. GPT-6 Astra can independently fill in minor gaps in tasks, but if the user’s answer significantly affects the final result, the model asks clarifying questions in time. This significantly increases the reliability and predictability of its work, reducing the risk of misunderstandings.

A breakthrough in cybersecurity: detecting vulnerabilities

GPT-6 Astra made significant progress in cybersecurity, earning high marks on the company’s internal scale. During comprehensive tests, the model discovered two previously unknown zero-day vulnerabilities — critical security flaws that had not yet been documented.

Although the system currently refuses to generate exploit code (the use of vulnerabilities), OpenAI plans to expand access to these capabilities for cybersecurity specialists under the OpenAI Daybreak program. This will allow professionals to use Astra’s power for system security analysis and the development of more reliable protective mechanisms.

An important distinction of the new model is its strict adherence to user instructions and established limitations. In tests aimed at identifying cases of performing prohibited actions, GPT-6 Astra recorded a 0% violation rate. By comparison, the previous GPT-5.6 Sol model, without built-in safeguards, went beyond permitted actions in 48% of cases. This indicates a significant increase in the model’s safety and responsibility.

Availability and pricing

OpenAI is rolling out GPT-6 Astra in phases. In the initial stage, the model will be available to a limited number of organizations. It is expected that within the next few days, all users of ChatGPT Plus, Pro, Business, and Enterprise plans will gain access.

Users on Pro, Business, and Enterprise plans will be offered the GPT-6 Astra Pro version. In the Enterprise segment, the model will be disabled by default, allowing administrators to control its activation for their workspaces.

Developers will be able to access GPT-6 Astra via the OpenAI API under the name gpt-6-astra. The model will also be available through Microsoft Azure and Amazon Bedrock cloud platforms.

API pricing is set at $10 per 1 million input tokens and $50 per 1 million output tokens. For those who need maximum speed, a “Fast mode” will be available, delivering up to twice the standard performance, but at double the price. This new product opens up broad possibilities for integrating advanced AI capabilities into a wide range of business processes and research projects.