Supported models

SandBase's model catalog changes frequently. Use the live catalog or the generated Model API Reference instead of relying on a static model list.

The live catalog is the availability source of truth. A newly enabled model can appear there before its generated reference page is published; use the catalog response to confirm availability, then follow the model-specific page once it becomes available for the complete request schema.

Current source of truth

Browse Models for current availability and pricing. For production integrations, call GET /v1/models immediately before selecting or validating a model.

Browse by capability

Category	API reference	Primary API
Language, reasoning, vision, and tool use	LLM models	POST /v1/chat/completions
Claude-compatible messages	Anthropic models	POST /v1/messages
Image generation and editing	Image generation	POST /v1/run or OpenAI Images
Video generation and editing	Video generation	POST /v1/run
Audio generation and processing	Audio generation	POST /v1/run

Each generated model page contains the exact public model identifier, request fields, supported values, capabilities, and a working request. Disabled or internal-only registry entries are not published there.

List models through the API

curl https://api.sandbase.ai/v1/models \
  -H "Authorization: Bearer $SANDBASE_API_KEY"

The response follows the OpenAI-compatible model-list format and contains enabled logical models available to your account. Treat returned model IDs as opaque strings and send them unchanged in the relevant request's model field.

Pricing and availability

Do not copy prices from a documentation table into application logic. Pricing, provider availability, routing, and enabled status can change independently of this page.

• Check live model pricing before estimating production cost.
• For a capability that returns a task ID (including an x-task-id response header), use GET /v1/tasks/{task_id}/cost to inspect settlement and usage. This is a cost lookup, not a generation-result endpoint: for asynchronous POST /v1/run calls, poll GET /v1/run/{id} with the returned opaque run ID instead.
• Handle model-unavailable and admission errors even when a model appeared in an earlier catalog response.
• Use the model-specific reference page rather than assuming every model supports the same context window, tools, vision, streaming, or reasoning parameters.

Choosing a model

Choose from the current catalog based on the operation you need, then verify the exact schema on its generated reference page. For a stable production integration, pin a specific logical model ID when available; use moving aliases only when you intentionally accept upstream model changes.

Check operational incidents separately on the SandBase Status Page.