OpenAI has launched GPT-6 Astra, a new artificial intelligence model focused on tasks involving computer use, programming, research, and data analysis. Unlike systems that only respond to commands, Astra can interact directly with websites and software to execute certain steps of a task.
With new context support, the model preserves debugging notes without losing historical data – Image: Samuel Boivin/Shutterstock
GPT-6 Astra is being released gradually and will be available to users of the ChatGPT Plus, Pro, Business, and Enterprise plans. GPT-6 Astra is being released gradually and will be available to users of the ChatGPT Plus, Pro, Business, and Enterprise plans.
In addition to knowing who can already access the model, it is also useful to understand how to assign a task to Astra and in which situations it may ask for additional information before proceeding.How to use the GPT-6 Astra?
The GPT-6 Astra was developed to perform tasks beyond answering questions. Instead of merely indicating how an activity should be done, the model can interact with websites, apps, and other programs to carry out different steps of the work. To use this feature, the user can describe the goal they want to achieve and provide the necessary context for the task, including the materials to be used and the expected result.
In practice, a request can combine several actions. OpenAI cites, for example, tasks such as researching information on the internet and preparing summaries in document editors, as well as analyzing data and generating charts. The model can also work with documents, spreadsheets, and presentations.
Who can already use the GPT-6 Astra?
The GPT-6 Astra is being released gradually. According to OpenAI, in the coming days the model will be available for users of the ChatGPT Plus, Pro, Business, and Enterprise plans. The Pro, Business, and Enterprise plans will also have access to Astra Pro. In the case of Enterprise accounts, the administrator needs to enable the feature.
In addition to ChatGPT, the model can also be accessed via the OpenAI API, under the name "gpt-6-astra". For plans that include the model, the use of Astra is part of the subscription limits, although additional credits can be purchased when necessary.
How to give a task to Astra?
One way to make use of the resource is to think less about a question and more about a task that can be delegated. The user can inform what needs to be done, which files or sources should be considered, which conditions must be respected, and what result they expect to receive.
For example, in a research task, it is possible to ask Astra to search for information on the internet and organize the main data found. In a task involving a spreadsheet, the request can indicate which information should be analyzed and ask for the results to be presented in charts or in a presentation.
When can Astra ask for more information?
It is not always necessary to determine each step of the task. OpenAI states that when an instruction leaves some gaps, Astra can use the context to handle routine situations. However, when missing information could change the outcome of the activity, the model may ask a specific question before proceeding.
In tasks that involve important decisions, Astra can also wait for confirmation before moving forward. According to OpenAI, the model avoids simply assuming an answer when the decision is consequential and may ask for clarifications to continue execution.
What can you do with Astra?
Using a computer is one of the areas in which the model was developed to act. From a instruction, it can navigate through pages, fill fields, and interact with different screen elements.
Among the examples presented are updating records in CRM systems, organizing calendars, and conducting online research. The model can also prepare summaries in text editors and work with information found during navigation.
The capability is not limited to the browser. Astra was tested in engineering and design software, including tasks in KiCad and generating CAD code for three-dimensional objects.
Documents, websites, and other projects
The model can also combine different tools during the same task. Some examples include:
- Presentations: creating slides from models and patterns provided by the user;
- Spreadsheets and documents: analyzing data, producing charts, and formatting files;
- Engineering and design: generating CAD code and working with circuit board projects;
- Websites and games: creating pages, applications, and small games from instructions;
- 3D modeling: working in Blender and transforming the result into an explorable scene in Unreal Engine 5.
In a test cited by OpenAI, Astra modeled a house in Blender and used it to create a scene in Unreal Engine 5.
Silas Alberti, senior vice president of research at Cognition, the company responsible for Devin, stated that the integration of the model improved tasks involving computer use, writing, and code comprehension. According to him, the videos produced during the tests became easier to follow and the reports, clearer and more concise.
Digital routine automation eliminates the need to perform each click manually. Image: One Artist/Shutterstock
Programming and scientific research
For those working in programming, Astra can perform tasks in the terminal, test systems, install software, and help in problem investigation. In Codex, there is also an experimental feature to preserve information between different context windows, which can be useful in longer projects.
Read more:
- GPT-6 Astra arrives and OpenAI says that the AGI journey may have started
- What is AGI and why OpenAI says we might have arrived there
- OpenAI robots hijack site and create secret forum on the web
In science, the model can operate specialized software and handle research data. Examples include analyzing the quality of genetic sequencing, visualizing variations, and evaluating evidence to guide new investigations.
The model also attempts to handle tasks that change during execution. When information is needed to decide between different paths, it can ask for clarification. In less important situations, it can proceed with a reasonable assumption.
The more steps depend on specific decisions, the more important it is to make those conditions clear.
The proposal, therefore, is less about an AI that merely answers questions and more about a tool capable of executing certain parts of a task. The extent to which this reduces the user's effort depends, in practice, on the type of task and the tools involved.
Valdir Antonelli is a journalist with a specialization in digital marketing and consumer behavior.
Ana Figueiredo is the technology editor at Olhar Digital. She is a journalism graduate from the Federal University of Uberlândia (UFU).
