Model catalogue
We expose one canonical ID per model. Under the hood, each model may be served by multiple providers. You don’t pick — the router does.
- Canonical IDs unify pricing and coverage across providers
- Tier keywords (
light,medium,high) let you ask for quality level instead of a specific model - Live listing at
/v1/models- always the current source of truth - Roadmap models are loaded as they come online; the catalogue is not static
Canonical IDs
A canonical ID looks like llama-3.3-70b or qwen-3.5-27b. It is stable — if a new provider starts serving the same model next week, the ID doesn’t change, only the available routes do.
When you send a request, we look up the canonical ID, find every provider who can serve it, filter and score them, and pick a winner. If you pass an alias instead (a date-stamped or vendor-specific string), we map it to the canonical ID behind the scenes.
Listing what’s available
The source of truth is always the API.
curl https://smart.aivory.net/v1/models \
-H "Authorization: Bearer $AIVORY_API_KEY"
The response includes every canonical ID plus an availability field:
| Availability | Meaning |
|---|---|
instant |
Served by hosted API providers or an active spot GPU. Requests are handled immediately. |
deploying |
A spot GPU is being provisioned for this model. estimated_deploy_seconds shows the ETA. |
cold |
Can be deployed on a spot GPU. The first request triggers deployment (3-10 min) and places a credit hold for the minimum session cost. |
pricing_pending |
Model is available from a provider but pricing has not been confirmed yet. Requests may route to it once pricing is set. |
A low provider_count means fewer fallback options for that model.
Tier keywords
If you don’t care which specific model runs, pass a tier keyword:
light- cheap, fast, good for summarisation, classification, simple routingmedium- general-purpose assistant workhigh- long-context reasoning, frontier quality
The router picks the cheapest model in that tier for your request. You get the canonical ID that actually served you back in the response’s model field.
Capabilities
Each model entry in /v1/pricing?format=flat includes a capabilities block:
{
"context_window": 131072,
"max_output_tokens": 4096,
"supports_function_calling": true,
"supports_json_mode": true,
"quality_tier": 2
}
If you pass tools to a model whose provider doesn’t support function calling, the router filters that provider out. If no provider can serve the request with the required capability, you get a 404 model_not_found.
Roadmap vs live
We publish models as they become live. “Soon” means nothing on this page; check /v1/models for what’s actually serving traffic right now. If you need something that isn’t there, write to support - we may already be working on adding it.
Browse model pages
We currently serve 16 open-weight models across families like Llama, DeepSeek, Qwen, Gemma, Mistral, and Voxtral. Each model has a dedicated page with live pricing, code examples, and use cases:
Browse all models with live pricing →
The catalogue is growing. If you need a model that isn’t listed, write to support.
Pricing per model
See /v1/pricing for the live rate card. The model index shows per-model pricing, and the main landing page’s live pricing table covers the full catalogue.