Models
The Models page is where you register, edit, and remove model definitions. Each model maps a model ID to a provider and defines its capabilities and pricing, which the gateway uses for routing, cost tracking, and budget enforcement.
Model List
Models are displayed in a paginated, searchable data table with the following columns:
| Column | Description |
|---|---|
| Model ID | The identifier clients use in requests (e.g. gpt-4-turbo, claude-3-opus) |
| Provider | The provider that serves this model |
| Capabilities | Tags indicating what the model supports |
| Context Window | Maximum token context length |
| Pricing | Input and output cost per million tokens (inputPerMillionTokens / outputPerMillionTokens) |
| Status | active, inactive, or deprecated |
Use the search bar to filter by model ID or provider name. The table supports sorting on Model ID, Provider, Context Window, and Status.
Adding a Model
Click Add Model and fill in the form:
- Model ID — The identifier that clients will reference (e.g.
gpt-4-turbo). This cannot be changed after creation. - Provider — Select from the list of configured providers.
- Context Window — Maximum token context length (defaults to 4096).
- Status — Set to
activeto make the model available immediately. - Input Price — Cost per million input tokens. (Pricing is stored on the model as
pricing.inputPerMillionTokens, matching the built-in catalog in@ai-gateway/shared.) - Output Price — Cost per million output tokens (
pricing.outputPerMillionTokens). - Capabilities — Check all that apply.
Capabilities
Each model can declare one or more capabilities, which the gateway uses for routing decisions:
| Capability | Description |
|---|---|
chat | Conversational completions (messages API) |
completion | Text completions |
embedding | Vector embeddings |
image | Image generation |
audio | Audio transcription, translation, or speech |
function_calling | Tool/function call support |
vision | Image understanding in prompts |
streaming | Server-sent event streaming |
These tags drive cost tracking, model intelligence, and display. (The built-in model catalog in @ai-gateway/shared carries a richer tag set — e.g. prompt-caching, reasoning, code-execution, search-grounding, computer-use — in addition to the ones selectable here.) Accurate tagging keeps cost and routing decisions correct.
Editing a Model
Click Edit on any model row to open the edit modal. All fields except Model ID can be changed. Common operations include:
- Updating pricing when providers change their rates
- Marking a model as
deprecatedto stop new routing while honoring in-flight requests - Adding or removing capabilities
Deleting a Model
Click Delete to remove a model. A confirmation dialog warns that existing routing rules referencing this model may be affected. Deletion is permanent.
Pricing and Cost Tracking
Model pricing feeds directly into the gateway’s cost calculations. When a request completes, the gateway multiplies the token counts by the model’s configured input and output prices to compute the cost. This data appears in:
- Dashboard cost breakdowns
- Budget enforcement checks
- Usage reports
Keep pricing current to ensure accurate budget tracking and cost-optimized routing decisions.
Model Statuses
| Status | Effect |
|---|---|
active | Model is available for routing |
inactive | Model is hidden from routing but retained in configuration |
deprecated | Model is marked for phase-out; routing strategies may deprioritize it |