Puck AIModel Configuration

Model Configuration

By default, Puck AI uses models chosen based on their balance of cost and performance.

You can specify a different model, fine-tune the settings, or (on Launch plans and above) bring your own provider key.

See model availability for supported models.

The default models

Puck AI chooses a model based on the mode of each request:

Mode	Default model
Assembly	`openai/gpt-5.4-mini`
Design	`openai/gpt-5.6-luna`

On pay-as-you-go plans, token consumption is charged at the provider’s list price plus a 20% markup. For the default models:

You can top-up your Puck Cloud account any time, or specify auto-reload to ensure you never run out of credit.

To use a different model, set the model parameter in the puckHandler:

const handler = puckHandler(request, {
  ai: {
    model: "openai/gpt-5.4-nano",
  },
});

To use a different model for design-mode requests, set designMode.model:

const handler = puckHandler(request, {
  ai: {
    designMode: {
      allowed: true,
      model: "openai/gpt-5.6-sol",
    },
  },
});

You can also specify a model in the generate and chat APIs.

Use the providerOptions parameter to fine-tune model behavior, such as to control reasoning effort:

const handler = puckHandler(request, {
  ai: {
    providerOptions: {
      openai: {
        reasoningEffort: "low",
      },
    },
  },
});

Use designMode.providerOptions to fine-tune design-mode requests independently.

Requires the Launch plan or above

To run Puck AI on your own provider account, provide a providerApiKey alongside the model.

BYOK requests do not consume Puck AI credits. Your provider bills the model usage directly to your account.

const handler = puckHandler(request, {
  ai: {
    model: "openai/gpt-5.5",
    providerApiKey: process.env.MY_OPENAI_KEY,
  },
});

Requests made with your own key are still routed via Puck’s servers. For full self-hosting, get in touch.

We currently support the following models: