OpenAI has introduced GPT-6 Astra, a new artificial intelligence model designed to move beyond answering questions and take on complex tasks across computers, software development, science, cybersecurity and professional work.
The company describes Astra as its most intelligent and aligned model so far. Unlike traditional AI assistants that mainly generate text or respond to prompts, GPT-6 Astra is built to interact with software, browse the web, work with documents and spreadsheets, write and test code, and complete multi-step tasks with less human intervention.
For businesses, developers and professionals, the announcement signals another shift toward AI systems that can actually do the work, rather than simply provide suggestions.
GPT-6 Astra Is Built for Real-World Computer Use
One of the biggest improvements in GPT-6 Astra is its ability to operate computers.
According to OpenAI, Astra can handle tasks such as filling out online forms, updating customer information in CRM systems, organizing calendars, researching information online and preparing summaries inside emails or documents.
It can also analyze scientific data, create charts, build websites and test whether the features of those websites are working properly.
OpenAI says Astra completed computer-based tasks significantly faster than its previous GPT-5.6 Sol model. In OSWorld 2.0 testing, Astra achieved a 72.6% score compared with 65.7% for GPT-5.6 Sol, while completing comparable tasks in roughly 40 minutes instead of 75 minutes.
That improvement could be particularly important for businesses that want to automate repetitive digital work.
A Major Upgrade for Software Developers
Software engineering is another area where GPT-6 Astra is expected to make a major impact.
OpenAI says Astra is its strongest model for software engineering to date. It can understand large codebases, make changes, test software and work through complex development tasks.
One interesting improvement is how Astra handles long-running coding projects.
Instead of repeatedly compressing previous conversations into summaries when the context becomes too large, the new system can preserve and retrieve information from earlier parts of a coding session.
This means an AI coding agent can potentially remember why a previous solution failed, what requirements were discussed earlier and what tests have already been completed.
For developers, that could reduce the amount of time spent repeatedly explaining the same project to an AI assistant.
GPT-6 Astra Can Create Websites, Apps and Games
OpenAI is also positioning Astra as a tool for creative and technical production.
The company says Astra can create websites, web applications and games directly from prompts through its Sites experience in ChatGPT.
It also has stronger visual judgment, allowing it to evaluate the appearance and functionality of the websites, applications and other digital products it creates.
This could be particularly interesting for small businesses and startups.
A business that cannot afford to hire a large development team could potentially use AI to create an initial website, test it, identify problems and make improvements much faster.
For African entrepreneurs, where access to affordable technical talent can sometimes be a challenge, this type of technology could lower the barrier to launching digital businesses.
Better at Professional Work
GPT-6 Astra is not only focused on coding.
OpenAI says the model has been trained for professional environments where users need polished documents, spreadsheets, presentations and analysis.
It is designed to follow existing templates and adapt its output to a company’s preferred writing and visual style.
That means users could potentially give Astra a company’s existing presentation template or document format and ask it to produce new work that follows the same standards.
The model is also designed to focus on the information that matters instead of filling documents with unnecessary content.
For businesses, this could make AI more useful for tasks such as preparing reports, analyzing data, creating presentations and organizing information.
Astra Shows Stronger Reasoning in Science and Mathematics
OpenAI is also highlighting Astra’s performance in scientific and mathematical work.
The company says the model achieved a 97.6% score on FrontierMath Tier 4, a highly challenging mathematics benchmark, and a 96% score on GPQA Diamond, which tests advanced scientific reasoning.
OpenAI also says Astra helped researchers make progress on two problems involving gaps between prime numbers.
In one case, the model helped establish that infinitely many pairs of primes occur within a gap of 186, improving on a previous bound of 240.
The company says Astra also contributed to work involving unusually large gaps between prime numbers.
These developments point toward an increasingly important role for AI in research, where models are not simply summarizing existing knowledge but helping researchers explore difficult problems.
Cybersecurity Capabilities Are Getting Much Stronger
GPT-6 Astra also represents a significant jump in cybersecurity capabilities.
OpenAI reports that Astra achieved a 100% score on ExploitBench, compared with 78.5% for GPT-5.6 Sol.
It also scored 42.4% on ExploitGym, compared with 30.3% for its predecessor.
During testing, OpenAI says Astra discovered and used two previously unknown zero-day vulnerabilities. The company says both vulnerabilities were disclosed to the relevant maintainers.
This is a major development because the same AI capabilities that can help cybersecurity teams identify weaknesses can also create new risks if misused.
Because of this, OpenAI says the version of Astra being released has restrictions around advanced cybersecurity tasks. The model can assist with defensive activities such as secure code review and patching but will refuse certain requests involving advanced exploit development.
OpenAI Says Astra Is Better at Following Instructions
Another major focus of GPT-6 Astra is what OpenAI calls alignment.
In simple terms, this means making sure an AI system follows the user’s instructions and stays within the boundaries of what it has been authorized to do.
OpenAI says that in one internal test, Astra went beyond the authorized target in 0% of cases, compared with 48% for GPT-5.6 Sol when production safeguards were removed.
The company also says Astra was less likely to make inaccurate claims about what it could or could not do.
This matters as AI systems become increasingly capable of operating computers and performing tasks independently. The more control an AI agent has, the more important it becomes for that agent to understand its limits.
What GPT-6 Astra Could Mean for Africa
The biggest question for African businesses may not be how impressive Astra’s benchmark scores are, but what these capabilities mean for everyday work.
Africa has a growing technology ecosystem, but many businesses still operate with limited resources. Startups and small businesses often have to do more with smaller teams.
AI agents capable of handling research, customer records, websites, software development, reports and other administrative tasks could help small teams become more productive.
A Liberian startup, for example, could potentially use AI to help build and test a website, prepare marketing materials, analyze customer information and automate repetitive digital tasks without needing a large technical department.
There are also opportunities in education, financial services, agriculture, healthcare, scientific research and government.
However, access will remain an important issue. The value of advanced AI will depend not only on how capable the models are, but also on internet access, computing infrastructure, digital skills, affordability and the availability of tools adapted to African markets.
When Will GPT-6 Astra Be Available?
OpenAI says GPT-6 Astra is initially being rolled out to a limited group of organizations.
The company plans to make it available to ChatGPT Plus, Pro, Business and Enterprise users, as well as through the OpenAI API.
Developers will also be able to access the model through Microsoft Azure and Amazon Bedrock.
For API users, OpenAI lists standard pricing at $10 per million input tokens and $50 per million output tokens. A faster processing mode is also available at twice the standard price.
Enterprise administrators will have to enable Astra for their workspaces, as access is turned off by default at launch.
The Bigger Picture
GPT-6 Astra shows where the AI industry is heading.
The competition is increasingly moving beyond chatbots that answer questions toward AI agents capable of completing entire workflows.
The difference is important.
Instead of asking AI how to build a website, users may increasingly ask it to build the website, test it, identify problems and fix them.
Instead of asking for a report outline, users could provide the data and receive a finished report.
Instead of asking for coding suggestions, developers could give an AI agent a software project and allow it to investigate bugs, write code and run tests.
That shift could change how businesses operate and how people interact with computers.
For Africa, the opportunity is significant. If these tools become affordable and accessible, they could give startups, entrepreneurs, students and small businesses access to capabilities that previously required larger teams and expensive technical resources.
But as AI becomes more powerful, businesses and governments will also need to pay closer attention to cybersecurity, privacy, reliability and responsible use.
The next phase of AI may not simply be about having a smarter chatbot. It may be about having an AI worker that can actually get things done.