Task requirements and reference materials
Perform numerical calculations
Still need to check against the goal
Model: A core system that is trained, processes inputs, and generates results, typically called by an application.
Why Applications and Models Get Mixed Up
In daily use, you usually only see one interface. The model that processes inputs behind the scenes is bundled with the features on the interface, so it's easy to treat "understanding requests" and "operating a computer" as the same ability.
Models can generate text, code, or other results based on inputs. An actual operation, such as reading directories, creating files, or sending notifications, usually still requires the application to connect to the corresponding tools and execute them.
A "Saved" Message Doesn't Tell the Whole Story
Suppose the model responds: "Lesson plan saved to file." Without an actual file or a successful tool result, this is still just text. To determine whether a file exists, you should open or retrieve the corresponding file, rather than relying on the statement.
| What Needs to Be Judged | What Evidence to Look For |
|---|---|
| Whether content meets requirements | Compare against student level, class duration, objectives, and materials |
| Whether a file was created | An actually openable file or a clear tool result |
| Whether visuals are normal | Actual page, preview, or rendered image |
Choose Capabilities Based on Task Requirements
Some tasks only need text processing, while others require reading images, understanding tables, or handling audio. Whether corresponding input/output is supported should be verified against the actual conditions of the model and application—avoid extrapolating capabilities across products.
The same model may appear differently in different applications: the context, tools, and settings it sees can vary, and so can the results. Changes in model versions may affect performance, so record the applicable environment for any operational examples.
Training and Current Reference Information Are Different
Training gives the model the ability to process inputs; providing reference materials for the current task is not the same as retraining on the spot. Loading a Skill typically means providing methods and materials.
That a model knows some general information doesn't mean it knows your class situation, the contents of your private files, or the latest changes. The specific information you need should still be provided and verified.
Next Steps
Read Context to understand what it can reference this time; or see how capabilities are combined in the PPT Case.
Think About It
The model writes "File saved." Can you conclude the file exists based on this?
Reference judgment: No—you need an actual file or a tool execution result; a text statement cannot replace execution.
Sources and Scope of Application
Reference OpenAI · Text generation, OpenAI · Agents SDK. Technical accuracy verified on 2026-09-09; cases and teaching organization are authored by this site. Specific product support scope and operating methods should be verified separately; this article makes no cross-product feature commitments.