OpenAI has introduced GPT-6 Astra, a new artificial intelligence model designed to assist with computer-intensive activities such as programming, research, and data analysis. Unlike systems that only provide answers to commands, Astra has the capability to interact directly with software and websites to complete specific steps in a process.
In practice, this functionality allows the AI to be used in various work tasks, such as filling out forms, updating records, organizing information, searching the internet, and manipulating documents, spreadsheets, and specialized programs.
One of the development focuses of the model is its operation within the computational environment. Based on an instruction, it can navigate web pages, enter data into fields, and interact with various visual elements on the screen. Practical examples include modifying records in CRM systems, managing schedules, and conducting online research. Furthermore, the model can generate summaries in text editors and process data collected during browsing.
Astra's capabilities extend beyond the browser; it has been tested in engineering and design software, performing operations in KiCad and generating CAD code for three-dimensional models. The model also demonstrates the ability to integrate multiple tools within a single activity. One case cited by OpenAI shows Astra modeling a residence in Blender and subsequently using that model to create a scene in Unreal Engine 5.
Silas Alberti, Senior Vice President of Research at Cognition, the developer of Devin, commented that the incorporation of this model has enhanced tasks related to computer use, writing, and code comprehension. He observed that the videos generated during testing became easier to follow and the reports presented, clearer and more concise.
For programmers, Astra offers support in executing terminal commands, system testing, software installation, and assisting in fault identification. Additionally, Codex features an experimental resource aimed at maintaining consistent information across different context windows, which can be beneficial in longer projects.
Scientific research and decision-making
In the scientific field, the model is capable of operating specific software and handling research data. This ranges from analyzing the quality of genetic sequencing to visualizing variations and evaluating evidence to guide new investigations.
Astra is also being trained to handle dynamic processes, where information needs may change during execution. If a decision must be made between multiple paths, the model has the capacity to request clarification. In less critical scenarios, it can proceed based on a considered reasonable assumption.
To optimize the use of Astra, the simplest recommendation is to clearly define the objective, available resources, and desired outcome. The more dependent the specific decision steps are, the more crucial it is to detail these conditions. The general proposal is to position AI not just as a question responder, but as a tool capable of executing concrete parts of a job, where the reduction of effort for the user will depend on the type of task and the tools employed.