OpenAI has just unveiled GPT-6 Astra, and this isn’t being positioned as a routine model update.
The company says Astra is its most capable model yet, with major improvements in coding, research, computer use, cybersecurity and complex multi-step tasks. More importantly, OpenAI is designing it to do more than simply answer questions — it can actually carry out work across software, websites and computer interfaces.
And that’s what makes GPT-6 Astra particularly interesting.
GPT-6 Astra Is Designed to Actually Do Things
Previous generations of ChatGPT became increasingly good at understanding instructions and generating answers.
Astra takes the idea further.
OpenAI says it can interact with computers to perform tasks such as filling out forms, updating CRM records, organizing calendars, conducting online research, creating websites and testing software. It can also work with documents, spreadsheets and presentations while following existing templates and instructions.
In other words, instead of telling you how to do something, Astra is increasingly capable of doing the work itself.
Codex Is Getting a Serious Boost Too
Developers are likely to notice one of the biggest changes in Codex.
OpenAI says Astra is its best model for software engineering to date, with improvements in coding, reasoning and long-running development tasks. Astra can also preserve information across context windows in Codex instead of relying entirely on repeatedly compressing previous work into summaries.
That’s important for large projects.
When an AI spends hours debugging a complicated application, losing small details from earlier attempts can cause it to repeat mistakes. Astra’s new context-preservation approach is designed to retain those details and make them searchable later.
Codex Is Also Getting Faster
OpenAI is updating the Codex harness alongside Astra.
According to OpenAI, combining the new model with the updated Codex environment results in 1.9× faster task completion compared with the current GPT-5.6 Sol experience on its Mind2Web benchmark.
That could make a noticeable difference for developers using AI agents for long-running tasks.
Instead of waiting for an AI to slowly work through a project, the goal is to have it navigate, code, test and troubleshoot much more quickly.
ChatGPT Could Become More Like an AI Employee
This might be the biggest change for everyday users.
OpenAI says Astra can handle multi-step workflows where the user provides a goal rather than a list of every individual instruction.
For example, the model can research information, interact with websites, organize information and produce a finished document or spreadsheet. It can also adjust its approach when the user changes requirements during a task.
OpenAI says Astra is better at knowing when to make a sensible assumption and when it needs to ask the user for clarification.
That’s a small-sounding improvement that could actually be extremely important for AI agents.
It Can Work Inside Your Computer
Astra’s computer-use capabilities are one of its biggest selling points.
OpenAI says the model can interact with browsers and applications to perform tasks such as:
- Filling out forms
- Researching products or services
- Updating records
- Working with spreadsheets
- Creating websites
- Testing websites
- Troubleshooting software
- Organizing information
- Performing professional workflows
OpenAI’s own testing showed Astra completing computer-use tasks faster than GPT-5.6 Sol, with the company reporting approximately 47% less time per task in its OSWorld 2.0 latency simulations.
And Then There’s the Cybersecurity Part
This is where GPT-6 Astra becomes much more controversial.
OpenAI says Astra is the first model to reach its Critical cybersecurity capability threshold.
The company says Astra can find previously unknown vulnerabilities and develop ways to exploit them across protected systems with limited human guidance. Because of that capability, OpenAI has introduced stronger safeguards and monitoring around the model.
That means Astra isn’t simply getting better at writing code.
It’s potentially getting much better at understanding how real-world computer systems work — including how they can be attacked.
OpenAI Is Also Claiming Major Safety Improvements
There’s another side to the story.
OpenAI says Astra is better at respecting user instructions and staying within defined boundaries. In one internal evaluation, the company reported that Astra went beyond the authorized target in 0% of cases, compared with 48% for GPT-5.6 Sol when tested without production safeguards.
OpenAI has also added additional monitoring that can detect situations where an AI agent may have misunderstood instructions. In supported scenarios, the system can pause or stop the conversation so the user can review what happened.
That matters because the more independently an AI can operate, the more important it becomes to make sure it doesn’t take the wrong action.
The Numbers Are Wild
OpenAI is reporting some eye-catching benchmark results for Astra.
The company says GPT-6 Astra scored:
- 99.9% on ARC-AGI-3
- 98% on FrontierMath Tier 4
- 100% on ExploitBench
OpenAI also says Astra has helped solve long-standing mathematical problems and represents a major improvement in computer-use capabilities.
Of course, benchmarks don’t necessarily tell you exactly how the model will behave in everyday use.
But they show why OpenAI is treating Astra as a major generational upgrade rather than another incremental release.
When Can You Use GPT-6 Astra?
This is the part many ChatGPT users will want to know.
OpenAI says GPT-6 Astra is initially rolling out to a limited group of organizations, with broader availability coming to ChatGPT Plus, Pro, Business and Enterprise users. It is also being made available through the OpenAI API, Microsoft Azure and Amazon Bedrock.
For developers, the API model is available as gpt-6-astra. OpenAI’s current API pricing is listed at $10 per million input tokens and $50 per million output tokens.
This Could Change What We Expect From ChatGPT
The most interesting thing about GPT-6 Astra isn’t simply that it’s smarter.
It’s that OpenAI is pushing ChatGPT and Codex toward becoming systems that can complete entire workflows.
The difference between:
“Tell me how to do this.”
and
“Go do this for me.”
could become increasingly small.
Astra is designed to research, browse, write, code, interact with software, test its own work and adapt when requirements change.
That’s a much bigger shift than simply getting better answers.
The Reality Check
Despite all the excitement, GPT-6 Astra doesn’t mean ChatGPT can magically do everything without supervision.
Its availability is still rolling out, some of its most powerful capabilities are restricted, and OpenAI itself is adding additional monitoring because increasingly autonomous AI systems introduce new risks.
But if Astra delivers what OpenAI is promising, GPT-6 could mark the point where ChatGPT starts feeling less like a chatbot and more like a digital worker.
And that’s probably the most important part of this entire release.
