Model Workbench Tutorial: Choose a Model, Run It, and Save the Result
2026-11-18 · 原创·功能教程
Using the Model Workbench: Choose a Model, Run It, and Save the Result
The Model Workbench puts model selection, running, and saving results on one path. It is aimed at users who are choosing a model, tuning parameters, or saving media results, including those working with a model for the first time.
What to prepare
Before opening anything, settle three points. First, the task type: text generation, dialogue, or a task that produces media results. Models differ in what they support, so the task decides the shortlist. Second, the input: for a first run, use a short sample that contains no sensitive information, then switch to real data once the flow works. Third, the budget boundary: pricing is shown on the model detail page, so review the billing basis before running rather than after.
Step 1: Open a model from the catalog
Go to the in-site catalog at /models and browse the available models. Pick the one that matches your task and open its detail page. Everything that follows happens from there.
Step 2: Review capabilities, pricing, and the model ID
Do not type anything yet. Check three items in order. Capabilities: confirm the supported input and output types cover your task. Pricing: read the billing basis and estimate the cost of this run. Model ID: this is the identifier you will place in code when calling the model, so copy it rather than retyping it.
Step 3: Enter your input and run
Fill in your content in the run area and execute. For a first attempt, keep the input short and check whether the returned result matches expectations. If it does not, refine the input and try again before switching models; input clarity usually matters more than model choice at this stage.
Step 4: Review and save the result
The result appears on the page after the run. If the output is a supported media result, save it to your own results library for later review and reuse; saved results stay private to you. For text output, rely only on what the page actually offers and do not assume export options that are not shown.
Step 5: Copy the sample code
The workbench provides sample code, and that code follows your current input: what you filled in on the page is reflected in the snippet, so a copy is a usable starting point for integration. Before copying, re-check the model ID and input fields against what the page displays.
Safety and cost notes
- Do not submit sensitive information. During early runs, avoid keys, personal identifiers, and unredacted business data.
- Start small. Validate the flow with a short input, then scale to real workloads.
- Watch the billing. Each run may incur cost, and repeated debugging adds up.
- Trust the page. Capabilities and pricing are those shown on the detail page; do not assume features that are not listed.
- Manage the output. Saved media results enter your own library, so confirm this fits your data handling requirements.
Next step
Once one full cycle is complete, return to /models to compare another model's capabilities and pricing against the first.