The model that builds your app and the model your app calls for its users are separate choices. Changing the builder's model picker does not rewrite an existing chatbot or other AI feature in your app.
Find what your feature uses
Ask the builder:
Inspect the AI request for this feature without changing it. Which endpoint and model does it use? Is the model explicit, or selected automatically?
For the built-in chat endpoint, a request can name a model. If it does not, the platform chooses one using the request's intent and the app's deployed subscription-tier configuration.
Change and test the model
Ask the builder to use a specific supported model for the feature. Check the result in Preview, including a failure case, before publishing. Test the published app too: it runs its deployed code, which may differ from your current working copy.
For developers, GET /api/ai/models on the app lists its advertised model IDs and configuration. The catalog includes Anthropic/Claude as well as other providers. The exact list and aliases depend on the deployed version and enabled capabilities; a catalog entry alone is not proof that a particular request will succeed. Use the actual endpoint and input your feature needs, and handle a provider refusal rather than silently substituting a different model.
Image generation, structured answers, vision and ordinary chat have different input and routing rules. Do not assume a chat model ID works on every AI endpoint.
Cost and exact checks
Built-in AI usage is charged to the workspace that owns the app. Model choice and measured usage affect the charge; check credit history after a small test. Buyer credits are a separate customer allowance, not a replacement for the workspace's AI bill.
Enforce exact rules, such as approved lists and numeric limits, in code. A model upgrade does not guarantee exact answers.