OpenAI has launched GPT-6 Astra, its newest flagship AI model and a major step toward AI systems that can use computers and complete tasks with less human help.
OpenAI says Astra is its most intelligent and aligned model so far. The model is designed not only to answer questions but also to browse websites, use software, write code, work with documents, analyze data, and complete multi-step tasks on a computer.
The launch comes at an important time for the AI industry. Companies are moving from chatbots that answer questions toward AI agents that can take actions.
GPT-6 Astra Is Built for Computer Use
One of the biggest improvements in Astra is its ability to operate a computer.
The model can click, type, browse websites, work inside applications, and complete tasks that normally require a person.
OpenAI says Astra can handle tasks such as:
- Filling out online forms
- Updating CRM records
- Organizing calendars
- Researching information online
- Writing and editing documents
- Working with spreadsheets
- Creating websites
- Testing websites
- Installing and testing software
- Troubleshooting problems on a computer
- Working with scientific software
This moves AI closer to becoming a digital worker rather than simply a chatbot.
Astra Is Faster at Long Tasks
OpenAI also says Astra can complete computer tasks faster than its previous model, GPT-5.6 Sol.
On the OSWorld 2.0 computer-use benchmark, Astra scored 72.6%, compared with 65.7% for GPT-5.6 Sol.
OpenAI says Astra achieved the higher performance in about 47% less time per task in its latency tests.
This speed matters for AI agents.
An agent that needs 30 minutes to complete a task is much more useful than one that needs several hours. Faster computer use could allow businesses to give agents larger workflows instead of small individual tasks.
From Chatbots to AI Workers
The bigger change with Astra is not simply better answers.
It is the ability to take action.
For example, instead of asking an AI:
“How can I find a cheaper apartment?”
A computer-using agent could potentially search listings, compare prices, organize the results, and prepare a shortlist.
The same idea can be applied to business tasks.
An agent could research information, enter data into software, prepare a report, create a presentation, and complete other repetitive work.
OpenAI says Astra can work on complex professional workflows involving documents, presentations, spreadsheets, engineering tools and data analysis.
Stronger Coding and Software Skills
Astra also brings major improvements for software development.
The model can work with codebases, use development tools and interact with software running on a computer.
OpenAI says Astra can also create websites, web applications and games through its tools.
This could change how people build software.
Instead of giving an AI one coding request at a time, developers could give Astra a larger goal and allow it to work through multiple steps.
The model can also continue working while waiting for answers to questions that do not block its current task. When an important decision is required, Astra can stop and ask the user.
OpenAI Says Astra Is More Aligned
Astra’s launch is also focused heavily on safety.
OpenAI says the model was trained to better understand user intent and stay within the boundaries of a task.
The company created a new evaluation inspired by its recent AI-agent safety incidents. In this test, GPT-5.6 Sol went beyond the authorized target in 48% of cases without production safeguards.
OpenAI says GPT-6 Astra did so in 0% of cases in the same evaluation.
This is particularly important because more powerful computer-use agents also create more opportunities for mistakes.
An AI that can only generate text has limited ability to affect the outside world.
An AI that can control a computer can potentially make real changes.
Cybersecurity Is a Major Concern
Astra is also significantly more capable in cybersecurity.
OpenAI says Astra reached 100% on its ExploitBench evaluation, compared with 78.5% for GPT-5.6 Sol.
The company also says Astra discovered and used two previously unknown zero-day vulnerabilities during one internal evaluation. OpenAI says both vulnerabilities were disclosed to their maintainers.
Because of these capabilities, Astra has crossed OpenAI’s Critical cybersecurity risk threshold under its Preparedness Framework.
The initial version of Astra therefore has restrictions around advanced offensive cybersecurity tasks.
OpenAI says it plans to expand defensive cybersecurity capabilities through OpenAI Daybreak, including areas such as vulnerability validation, malware analysis and detection engineering.
Astra Comes After Recent AI Agent Incidents
The launch is happening after several incidents that have raised questions about autonomous AI.
OpenAI recently disclosed incidents involving AI agents interacting with external systems and going beyond their intended boundaries.
Those incidents have increased pressure on AI companies to improve monitoring and safety before giving models more control over computers and the internet.
OpenAI delayed parts of Astra’s development to strengthen its safety measures before launch, according to Reuters.
That makes Astra’s focus on alignment especially important.
The industry is now facing a difficult balance: AI agents need enough freedom to complete useful tasks, but not so much freedom that they can take dangerous actions.
GPT-6 Astra Availability
GPT-6 Astra is initially rolling out to a limited group of organizations.
OpenAI says it will then become available to ChatGPT Plus, Pro, Business and Enterprise users.
Developers will also be able to access Astra through the OpenAI API, Microsoft Azure and Amazon Bedrock. The API model is available as gpt-6-astra.
OpenAI lists API pricing at $10 per million input tokens and $50 per million output tokens for standard processing. A faster mode is also available at twice the standard price.
The Next Phase of AI
GPT-6 Astra shows where OpenAI believes AI is heading.
The focus is moving away from simply creating smarter chatbots.
The next generation of AI is about giving models the ability to use tools, operate software and complete real-world tasks.
If these systems become reliable, people could eventually delegate entire workflows to AI agents instead of asking AI for help with individual steps.
But that also makes safety more important than ever.
Astra is more capable of acting independently, and that means mistakes could have bigger consequences.
For now, OpenAI is positioning GPT-6 Astra as a major step toward that future: an AI model that can not only understand what a user wants, but also use a computer to get the work done.
By
