OpenAI has officially launched GPT-6 Astra, describing it as its most capable and aligned model yet. The new model focuses heavily on computer use, coding, research, cybersecurity and complex multi-step work, with OpenAI positioning Astra as a system that can actually carry out tasks inside software rather than simply tell users how to do them.
The rollout has started with a limited group of organizations, while access for ChatGPT Plus, Pro, Business and Enterprise users, along with API access, is planned over the coming days.
What Is GPT-6 Astra?
GPT-6 Astra is OpenAI's newest frontier AI model and the successor to its GPT-5.6 generation.
The biggest change is the emphasis on end-to-end task completion. OpenAI says Astra can work with computers, browse websites, write and test software, analyze data, create documents and presentations, and handle complex professional workflows.
Instead of responding with a list of instructions, Astra is designed to perform more of the actual work.
OpenAI says it can fill online forms, update customer records, organize calendars, conduct online research, analyze scientific data, generate plots, build websites and perform frontend quality checks.
Why Is GPT-6 Astra Different From Earlier ChatGPT Models?
Earlier ChatGPT models became increasingly capable at reasoning, coding and tool use. Astra pushes that idea further by combining those abilities with stronger computer interaction.
For example, a user could give Astra a complicated task involving multiple applications. The model can work through the steps, adapt when requirements change and continue with parts of the job that do not require additional human input.
OpenAI says Astra is also better at recognizing when an instruction is ambiguous. It can make sensible assumptions for routine decisions while asking the user when a missing detail could materially change the outcome.
That could make AI agents considerably more useful for real workplace tasks.
What Can GPT-6 Astra Do?
OpenAI is highlighting several major areas where Astra improves over previous models.
1. Computer Use
Computer use is arguably the biggest part of the Astra launch.
The model can interact with software and websites to complete tasks such as:
- Filling online forms
- Updating CRM records
- Organizing calendars
- Conducting web research
- Creating and testing websites
- Running frontend QA
- Installing and testing software
- Troubleshooting issues visible on screen
OpenAI reports that on its OSWorld 2.0 evaluation, Astra scored 72.6%, compared with 65.7% for GPT-5.6 Sol. OpenAI also says Astra completed these simulated tasks in roughly 40 minutes compared with about 75 minutes for GPT-5.6 Sol in its latency simulation.
The company says Astra's computer-use performance translates into approximately 1.9× faster task completion than the current GPT-5.6 Sol experience on the Mind2Web benchmark when combined with an updated Codex harness.
2. Coding and Software Engineering
OpenAI calls GPT-6 Astra its best software-engineering model to date.
It is designed for agentic coding workflows in which the model can write code, execute it, test the result and iterate rather than stopping after generating a code snippet.
OpenAI's published benchmarks show Astra at 57.9% on Terminal-Bench 4.0, compared with 37.3% for GPT-5.6 Sol.
On DeepSWE v1.1, Astra scored 74.1%, ahead of GPT-5.6 Sol's 72.7% and Claude Opus 5's 73.7% in OpenAI's comparison.
This doesn't mean Astra will automatically produce perfect production code. Real-world software projects still require testing, review and human oversight.
How Good Is GPT-6 Astra at Research and Science?
Astra also represents a major push into scientific and mathematical work.
OpenAI reports a 97.6% score on FrontierMath Tier 4, compared with 83.0% for GPT-5.6 Sol in its evaluation table.
It also scored 96.0% on GPQA Diamond, compared with 94.6% for GPT-5.6 Sol.
OpenAI says Astra has already contributed to solving long-standing open mathematical problems. These claims should be understood as examples of the model's capabilities rather than evidence that AI has independently replaced mathematical researchers.
What About Documents, Spreadsheets and Presentations?
GPT-6 Astra is also designed for professional knowledge work.
OpenAI says it can create:
- Documents
- Spreadsheets
- Presentations
- Data analyses
- Reports
More importantly, Astra can follow existing templates and instructions and adapt when the user changes requirements.
The model is also designed to pay attention to the context that actually matters rather than unnecessarily repeating information. OpenAI says this should result in artifacts that are more immediately usable in professional environments.
For businesses, this could be one of the most practical improvements because many office tasks involve moving information between documents, spreadsheets, browsers and internal systems.
How Does GPT-6 Astra Perform Against Other AI Models?
OpenAI's own benchmark table places Astra ahead of several competing models in a number of evaluations.
| Benchmark | GPT-6 Astra | GPT-5.6 Sol | Claude Fable 5.1 | Claude Opus 5 | Gemini 3.8 Flash |
|---|---|---|---|---|---|
| OSWorld 2.0 | 72.6% | 65.7% | — | 70.2% | — |
| ScreenSpot-Pro | 92.7% | 76.9% | — | — | — |
| AutomationBench | 41.4% | 18.1% | 31.4% | 26.9% | — |
| BenchCAD | 95.9% | 83.3% | 84.3% | 82.1% | — |
| Terminal-Bench 4.0 | 57.9% | 37.3% | 55.8% | 52.3% | 19.1% |
| DeepSWE v1.1 | 74.1% | 72.7% | 67.4% | 73.7% | 73.8% |
| FrontierMath Tier 4 | 97.6% | 83.0% | 87.8% | 73.2% | — |
| GPQA Diamond | 96.0% | 94.6% | 93.7% | 93.7% | 95.3% |
These are OpenAI-reported evaluations, and the company notes that some tests were conducted in research or API environments. Production ChatGPT can differ because of system prompts, available tools and other deployment differences.
So the results show substantial gains in several areas, but they should not be interpreted as Astra being universally better at every possible task.
What Is GPT-6 Astra's Context Window?
The developer specification lists a 1,050,000-token context window and a maximum output of 128,000 tokens.
That gives Astra enough capacity to work with very large amounts of information within a single context, which can be particularly useful for large codebases, lengthy documents and research workflows.
The API documentation lists image input support, while audio and video are not currently supported as direct model modalities in the API specification.
How Much Does GPT-6 Astra Cost?
For developers using the API, OpenAI lists GPT-6 Astra at:
| API usage | Price per 1 million tokens |
|---|---|
| Input | $10 |
| Cached input | $1 |
| Cache writes | $12.50 |
| Output | $50 |
OpenAI also offers a Fast mode priced at twice the applicable standard rate, while Batch and Flex processing are priced at 50% of standard rates. Requests exceeding 272,000 input tokens are subject to higher pricing.
For ChatGPT users, access is tied to the staged rollout across eligible paid plans rather than a separate consumer “Astra” subscription.
When Will GPT-6 Astra Be Available in ChatGPT?
GPT-6 Astra is not yet generally available to everyone.
OpenAI says the initial rollout is going to a limited set of organizations. Broader access is planned for ChatGPT Plus, Pro, Business and Enterprise users over the coming days. API access is also rolling out.
The developer documentation similarly lists enterprise access through the Trusted Access Program first, followed by API and paid ChatGPT plans.
This means users should not assume that Astra will immediately appear in every ChatGPT account.
Why Is Cybersecurity One of the Biggest Astra Stories?
GPT-6 Astra isn't only a stronger coding model. OpenAI has classified it as its first model to reach the Critical cybersecurity capability threshold under its Preparedness Framework.
OpenAI says that with appropriate tools and access, Astra can discover previously unknown vulnerabilities and develop new exploitation techniques across well-protected systems without a person guiding every individual step.
That capability has obvious defensive applications, but it also creates greater misuse risks.
Because of this, OpenAI says it has strengthened protections against harmful cyber actions and added stricter isolation, monitoring and alignment evaluations.
Is GPT-6 Astra Safer Than GPT-5.6 Sol?
OpenAI says Astra is substantially more robust against jailbreaks and better at respecting safety and security boundaries than GPT-5.6 Sol.
One internal evaluation described by OpenAI tested whether a model would go beyond an authorized target. Without production safeguards, GPT-5.6 Sol exceeded the authorized scope in 48% of cases, while Astra did so in 0% of the company's test cases.
However, OpenAI also acknowledges an important limitation: Astra's monitorability has decreased compared with GPT-5.6 Sol.
The company says Astra is more capable of controlling its own chain-of-thought and less likely to include incriminating information there. That creates an ongoing challenge for researchers trying to understand and monitor increasingly capable models.
Does GPT-6 Astra Mean AGI Has Arrived?
This is likely to become one of the biggest debates following the launch.
OpenAI President Greg Brockman has described Astra as potentially marking the beginning of the AGI era. But AGI does not have one universally accepted definition or a single agreed-upon test.
Astra's ability to use computers, code, research and complete professional workflows certainly makes it more general-purpose than traditional chatbots.
Still, calling it definitive AGI would go beyond what can currently be established from benchmark scores and demonstrations.
The more important test will be whether Astra can reliably perform complex real-world work across different environments, users and situations without requiring constant human correction.
What Does GPT-6 Astra Mean for Everyday ChatGPT Users?
For normal users, the biggest change may be that ChatGPT becomes less like a question-and-answer tool and more like a digital worker.
Instead of asking:
“How do I organize this spreadsheet?”
Users could increasingly ask the model to organize the spreadsheet itself.
Instead of asking how to test a website, they could ask Astra to build the website, run tests and report what needs fixing.
Instead of asking for research sources and then manually compiling them, users could delegate more of the research workflow to the model.
That doesn't eliminate the need for human judgment. It changes where that judgment is needed—from manually completing every step to checking the model's decisions and final output.
Should You Upgrade to GPT-6 Astra?
For users who mainly ask simple questions, rewrite text or perform basic brainstorming, the difference may not immediately justify changing their workflow.
The bigger benefits are likely to appear for people doing:
- Software development
- Data analysis
- Research
- Business operations
- Document-heavy work
- Complex computer tasks
- Professional automation
- Long-running AI-agent workflows
For these users, Astra's stronger computer use and ability to handle multi-step tasks could be considerably more valuable than a simple improvement in chatbot answers.
FAQ
What is GPT-6 Astra?
GPT-6 Astra is OpenAI's latest flagship AI model, designed for complex reasoning, coding, computer use, research, cybersecurity and professional workflows.
Is GPT-6 Astra available now?
Yes, but access is currently rolling out in stages. OpenAI says broader access for Plus, Pro, Business and Enterprise users is planned over the coming days.
How much does GPT-6 Astra cost?
The API costs $10 per million input tokens and $50 per million output tokens under standard pricing. ChatGPT access depends on the eligible subscription plan and rollout status.
Is GPT-6 Astra better than GPT-5.6 Sol?
OpenAI's published benchmarks show Astra ahead of GPT-5.6 Sol across many computer-use, coding, science and cybersecurity evaluations. However, benchmark leadership does not mean it will be better at every real-world task.
The Bottom Line
GPT-6 Astra represents a significant shift in what OpenAI expects its models to do. Rather than simply generating answers, the model is designed to operate computers, complete multi-step workflows, write and test software, conduct research and produce finished professional work.
Its benchmark results are impressive, particularly in computer use, coding, mathematics and cybersecurity. But the more important development may be the move toward AI systems that can act independently inside real software environments.
That also makes safety more important. Astra's launch shows that the AI race is no longer only about which model can generate the smartest answer—it is increasingly about which model can complete the most useful work while remaining reliable, controllable and safe.