Models

fx uses the Vercel AI Gateway model catalog. The compiled default is zai/glm-5.2-fast.

Before authentication, the catalog can show public models. With a supported credential and selected Vercel team, it can also include models available to that team.

Browse the AI Gateway model catalog for available models, capabilities, providers, and pricing. See AI Gateway for service details.

Browse and select

List models outside the shell:

$fx models
$fx models --json

Inside fx, open the catalog or select by ID or search query:

$/models
$/model openai/gpt-5.4

The selected model and reasoning effort are saved as user defaults in ~/.fx/settings.json.

Selection order

The first available source wins:

  1. FX_MODEL for the current process
  2. a legacy workspace model override, if one still exists
  3. the user default in ~/.fx/settings.json
  4. the compiled default

Project .fx.json files cannot set model, effort, or fast mode. This prevents a repository from silently changing your model preferences.

$fx status

Use fx status or /status to confirm the effective model.

To change the Vercel team used for team-scoped models, run fx teams. See Change Vercel team.

Temporary overrides

Override one process without changing your saved default:

$FX_MODEL=openai/gpt-5.4 fx

ACP has an equivalent flag:

$fx acp --model openai/gpt-5.4

Fast mode

/fast toggles fast mode where the selected model supports it. The setting is saved as a user preference.

Controls depend on the model

Reasoning effort and fast mode appear only when the selected model supports them.