With GPT-6 Astra, OpenAI has introduced its most capable AI model to date. Astra began rolling out on 3 September 2026 and is already available via the OpenAI API. Access through ChatGPT is being introduced gradually, meaning availability depends on your subscription tier, workspace configuration and rollout status.
Unlike traditional language models, OpenAI positions Astra as the foundation for a new generation of autonomous AI agents. Rather than simply answering questions, it is designed to plan complex tasks, use tools and execute multi-step workflows independently.
From chatbot to AI agent
OpenAI describes Astra as a model built for professional knowledge work. It can operate computers and web browsers, carry out extensive research, develop software, analyse data, and create documents, spreadsheets and presentations.
The key difference from previous ChatGPT models lies in its ability to organise complete workflows autonomously. Astra can, for example, research information, edit files, complete forms, test software or combine multiple tools in order to achieve a specific objective.
This brings OpenAI another step closer to a digital workplace assistant that does more than provide support—it can carry out substantial tasks on its own.
Who can use Astra today?
The rollout is taking place in phases. Developers can already access Astra through the OpenAI API using the model name gpt-6-astra.
The model is also being introduced across ChatGPT. Plus, Pro, Business and Enterprise users are receiving access gradually. Within Enterprise workspaces, Astra must first be enabled by an administrator. If the model has not yet appeared in your model picker, it does not necessarily mean your subscription is excluded from the rollout.
The biggest technical advances
OpenAI highlights three major improvements.
First, Astra is designed to manage long-running workflows far more effectively. It can retain information throughout extended sessions and retrieve earlier context directly instead of relying solely on compressed summaries.
Second, it works much more accurately with existing templates. Presentations, spreadsheets and documents are expected to preserve the structure, layout and style of existing company templates far more reliably.
Third, Astra is more resilient when handling complex tasks. If instructions are unclear, it asks targeted follow-up questions. Where only minor information is missing, it makes sensible assumptions without losing sight of the original objective.
Major improvements in coding and computer control
According to OpenAI, Astra delivers significant advances in software engineering, computer operation, browser understanding and workflow automation.
The company reports substantial improvements over GPT-5.6 Sol, particularly in software development and agentic computer interaction. OpenAI also states that Astra not only completes tasks more successfully but frequently finishes them considerably faster.
These performance figures are based primarily on OpenAI’s own benchmark evaluations and should therefore be viewed as vendor-reported results.
A model with unprecedented safety measures
Its performance is only part of the story. Astra’s safety classification is just as noteworthy.
OpenAI has classified Astra as the first of its models to reach the highest level of its Preparedness Framework for cybersecurity capabilities. According to the company, Astra possesses capabilities that could present significant risks if misused.
As a result, it is subject to considerably stricter safeguards than previous models. Certain cybersecurity features are initially restricted, high-risk requests are consistently blocked, and suspicious workflows receive additional monitoring. Depending on the product, security-sensitive actions may even require human approval or be terminated entirely.
OpenAI says these measures are necessary because Astra is the first model whose capabilities require this level of precaution.
A powerful platform for developers
For developers, Astra offers an exceptionally large context window, making it particularly well suited to extensive codebases, large document collections and sophisticated agent workflows.
The model is already available through the OpenAI API and is expected to be offered through Microsoft Azure and AWS Bedrock as well.
What Astra means for businesses
For organisations, Astra represents another step beyond the traditional chatbot towards an autonomous digital colleague.
Instead of answering isolated questions, it can support complete business processes: analysing software, preparing documents, processing data and conducting complex research. This opens up significant new opportunities, particularly for knowledge-intensive work.
At the same time, OpenAI emphasises that Astra should continue to be supervised. Changes to production systems, security-sensitive actions and business-critical decisions should still be reviewed by humans.
A major milestone for agentic AI
With Astra, the emphasis shifts once again. Whereas previous generations focused primarily on delivering better answers, the new priority is enabling AI to take meaningful action.
The real breakthrough therefore lies not in individual benchmark scores but in the combination of advanced reasoning, computer control and tool use. Together, these capabilities are likely to shape the next generation of professional AI applications—provided the associated safety risks remain manageable.
