September 15, 2026
OpenAI launched GPT-6 Astra on September 3, 2026, introducing what the company describes as its most capable and aligned model to date. Astra is designed for complex reasoning, computer use, browsing, software engineering, cybersecurity, science, and professional work. Unlike a conventional chatbot that mainly responds to prompts, Astra is built to help you work through longer, multi-step tasks across browsers, code, documents, and professional software.
The launch represents a broader shift in how OpenAI is developing ChatGPT. The focus is no longer only on getting better answers. With Astra, you can give the model a goal, let it work through the steps needed to complete it, give it access to the right tools, and adjust the task when your requirements change.
OpenAI says you can use Astra for tasks such as online research, CRM updates, document creation, website development, software testing, data analysis, and troubleshooting problems on a computer screen. The model has also posted strong results across mathematics, science, computer-use, coding, and cybersecurity evaluations.
Here is what has changed with GPT-6 Astra, how you can use its computer-use capabilities, what the latest benchmark results show, and what the update means for you and your business.
What is GPT-6 Astra?
GPT-6 Astra is OpenAI’s flagship model for what it calls the hardest end-to-end work.
The model combines improvements in pre-training, reinforcement learning, and alignment. OpenAI says these improvements allow Astra to perform at the leading edge across computer use, browsing, software engineering, cybersecurity, science, and professional work.
The important word here is end-to-end.
Earlier AI systems could help you with individual parts of a task. You could ask an AI to write code, summarize research, create a spreadsheet formula, or draft an email. With Astra, you can connect several of those steps together.
For example, you can give Astra a research objective. It can search for information, work with the results, organize the findings, create an analysis, and produce a finished document instead of stopping after answering one question.
That makes Astra more useful when you need to complete an actual workflow rather than simply get an answer to a single prompt.
What can GPT-6 Astra do?
With Astra, you can handle several tasks that previously required different AI tools or a lot of manual work.
OpenAI says you can use Astra to:
- Fill out online forms
- Update customer information in CRM systems
- Organize calendars
- Conduct online research
- Draft summaries in email and document editors
- Analyze scientific data
- Generate plots
- Create websites
- Perform frontend quality checks
- Install and test software
- Troubleshoot problems visible on a computer screen
The important part is not that each of these tasks is completely new to AI. The bigger change is that you can bring several of them together in one workflow.
For example, if you are working on a website, you may need to research a topic, write content, code a page, open it in a browser, check how it looks, identify an issue, and test the result.
Instead of handling each step separately, you can give Astra the broader objective and let it work through the different stages.
You can also give Astra incomplete instructions. OpenAI says the model can use context to fill routine gaps, ask focused questions when missing information could change the result, and incorporate new requirements without losing track of the wider task.
How does Astra use a computer?
Computer use is one of the biggest changes introduced with Astra. Instead of only telling you what to do, Astra can interact with a computer environment through supported tools. This means you can give it a task that requires several actions and allow it to work through those actions.
The basic process looks something like this:
Understand the task → inspect the environment → take an action → check the result → continue or adjust.
For example, if you ask Astra to check a website, it may need to open the page, interact with different elements, identify an issue, make a change, and then check whether the change worked.
OpenAI’s API documentation lists computer use as a supported tool for Astra, alongside web search, file search, code interpreter, hosted shell, image generation, MCP, and other capabilities.
This means you can combine Astra’s reasoning with different tools instead of treating the answer itself as the end of the task.
Astra is also faster on computer-use tasks
OpenAI reports that Astra achieved 72.6% on OSWorld 2.0, compared with 65.7% for GPT-5.6 Sol.
The difference also shows up in task time. In OpenAI’s latency simulations, Astra took roughly 40 minutes per task, compared with about 75 minutes for GPT-5.6 Sol.
That works out to about 47% less time per task while achieving the higher score.
| Model | OSWorld 2.0 | Approx. task time |
| GPT-6 Astra | 72.6% | ~40 min |
| GPT-5.6 Sol | 65.7% | ~75 min |
This matters if you are using AI for tasks that involve several computer interactions. You are not only looking for a model that can complete the task. You also want it to complete the work without taking unnecessarily long.
What does Astra change for coding?
If you work with software, you can use Astra for more than generating individual pieces of code. OpenAI describes Astra as capable of handling complex software-engineering workflows across code, browsers, and professional software. You can also use it to install and test software and investigate problems.
The workflow can look more like:
write → run → inspect → identify the problem → modify → test again
rather than:
write code → return code
OpenAI reports a 57.9% score on Terminal-Bench 4.0, compared with 37.3% for GPT-5.6 Sol in its published comparison.
Astra also supports hosted shell and code interpreter tools through the API, giving you more ways to combine the model with execution environments.
Astra can work with documents, spreadsheets and presentations
You can also use Astra for professional documents and other business deliverables.
OpenAI says Astra can create documents, spreadsheets, and presentations while following templates and instructions. It can also adapt when you add requirements or change direction during the work.
This can be useful if your business already has a specific way of creating documents.
For example, you may have:
- A fixed presentation template
- A standard reporting format
- A preferred spreadsheet structure
- Brand-specific writing guidelines
- Internal documentation requirements
Instead of asking AI to create something completely from scratch, you can give Astra the existing structure and ask it to work within those requirements.
That can save time when consistency matters more than creating a completely new format.
OpenAI’s model guidance also highlights Astra’s ability to preserve the broader task when you introduce new instructions. So, if you change a requirement halfway through the work, you do not necessarily have to restart the entire workflow.
What do Astra’s benchmark results show?
OpenAI’s launch included results across several different types of evaluation.
| Evaluation | GPT-6 Astra result |
| ARC-AGI-3 | 99.9% |
| FrontierMath Tier 4 | 98% |
| GPQA Diamond | 96.0% |
| ExploitBench | 100% |
| OSWorld 2.0 | 72.6% |
| ScreenSpot-Pro | 92.7% |
| Terminal-Bench 4.0 | 57.9% |
| AutomationBench | 41.4% |
These tests measure different capabilities, so you should not treat them as one overall intelligence score.
For example, ARC-AGI-3 measures a different type of problem-solving ability from OSWorld, which focuses on computer interaction. Terminal-Bench evaluates software-engineering tasks, while ExploitBench focuses on cybersecurity capabilities.
Still, the results give you a useful picture of what OpenAI is trying to improve with Astra: reasoning combined with the ability to act on that reasoning.
Astra also achieved 88.0% on SRE-Bench in a single attempt and 99.2% within four attempts, according to OpenAI’s cybersecurity evaluation. GPT-5.6 Sol scored 55.9% and 68.7% respectively in the same comparison.
Astra is also built for research and science
You can use Astra for scientific and mathematical work that goes beyond simply summarizing research.
OpenAI reports that Astra reached 98% on FrontierMath Tier 4 and says the model has helped solve long-standing open problems in mathematics.
The model can also combine scientific reasoning with computer use. OpenAI describes workflows where Astra can work with specialized software, inspect scientific data, explore results, and help researchers determine what to investigate next. This matters because scientific work rarely ends with a written answer.
You may need to move between:
research papers → datasets → software → analysis → visualization → interpretation
A model that can work across those steps can be more useful than one that only explains the science to you.
Suggested image: Scientific data analysis or sequencing workflow screenshot from the official Astra announcement.
How large is Astra’s context window?
Astra has a 1,050,000-token context window and supports up to 128,000 output tokens through the API. OpenAI lists its knowledge cutoff as April 30, 2026. A context window of this size can be useful when you are working with large amounts of information.
That could include:
- Long technical documents
- Large research collections
- Complex software projects
- Extensive business documentation
- Large datasets and analysis workflows
Astra also supports several reasoning-effort levels: low, medium, high, xhigh, and max.
This gives you more control over how much reasoning effort is used for a particular task. A simple request may not need the highest setting, while a difficult research or engineering problem can use more reasoning.
What does Astra cost?
If you are using Astra through the API, OpenAI currently lists the following pricing:
- $10 per 1 million input tokens
- $1 per 1 million cached input tokens
- $12.50 per 1 million cache-write tokens
- $50 per 1 million output tokens
Requests containing more than 272,000 input tokens are priced at higher rates for the full request. OpenAI also lists separate pricing options for Batch, Flex, and Fast modes.
At first glance, Astra’s per-token pricing is higher than some of OpenAI’s other models. However, OpenAI says Astra can sometimes complete complex tasks using substantially fewer output tokens. So, if you are evaluating Astra for business use, looking only at the cost per token may not give you the complete picture.
A more useful question is:
How much does it cost to complete the entire task?
Why is safety especially important with Astra?
The more capable an AI model becomes at using computers and tools, the more important its safety controls become.
If a model can only generate text, its ability to directly affect an external system is limited. But if you give a model access to browsers, code execution, applications, and multi-step workflows, a mistake can have wider consequences.
OpenAI says Astra is its first model to reach the Critical level of cybersecurity capability under its Preparedness Framework.
The company says that, with the right tools and access, Astra can identify previously unknown security weaknesses and develop ways to exploit them without a person guiding every step.
During its evaluations, Astra also discovered and used two previously unknown zero-day vulnerabilities, which OpenAI says it disclosed to the relevant maintainers.
Because of these capabilities, OpenAI has added stronger safeguards, monitoring, and protections around computer-using workflows.
The company also says Astra is more resistant to prompt injection and safer in browsing and workplace environments than GPT-5.6 Sol.
Its production deployment includes additional misalignment monitoring, and certain actions may be paused or stopped when the system detects a potential safety issue.
So, while you can give Astra more control over computer-based tasks, that does not mean you should treat it as an unrestricted autonomous system. Permissions, monitoring, and human review remain important, particularly for sensitive business, financial, or security-related work.
Who can use GPT-6 Astra?
OpenAI announced Astra, initially rolling it out to a limited set of organizations. The company said broader access would follow for ChatGPT Plus, Pro, Business, and Enterprise users, as well as through the OpenAI API, Microsoft Azure, and AWS Bedrock.
Availability is being rolled out gradually, so your access can depend on the product, plan, and account you are using. OpenAI’s current Help Center also says GPT-6 Astra is being rolled out in ChatGPT Work and Codex, with availability varying between Chat, Work, and Codex during the rollout.
If you are a developer, Astra is available through the Responses API and Chat Completions, with tools including web search, file search, code interpreter, hosted shell, computer use, image generation, and MCP.
What does GPT-6 Astra mean for ChatGPT?
The biggest change you may notice with Astra is the direction ChatGPT is taking.
Earlier generations of AI assistants largely focused on producing information. Astra moves closer to a model that can work with that information and act on it.
You can ask Astra to research something, analyze information, write code, create a document, or complete a computer-based task. It can then use the appropriate tools and continue working through the task rather than stopping after its first response.
That does not mean every task can or should be fully automated. You still need human judgment for decisions involving financial, legal, security, or other high-impact consequences.
But the direction is becoming clearer.
AI is moving from answering individual prompts toward completing larger pieces of work. GPT-6 Astra is one of OpenAI’s clearest steps in that direction so far.
How Primotech Can Help You Put AI to Work
For businesses, using advanced AI models such as GPT-6 Astra involves more than simply choosing a powerful model. You also need the right strategy, integrations, workflows, and implementation to turn AI capabilities into practical business solutions.
At Primotech, you can get support with AI consulting, AI/ML development, automation, AI agents, LLM solutions, and custom AI applications. The team can help you identify where AI fits into your existing processes and develop solutions around your specific business requirements.
If you are exploring how GPT-6 Astra and other advanced AI models can support automation, software development, research, customer workflows, or business operations, Primotech can help you plan and implement an AI solution suited to your goals.
