CrossModel

Model aliases

A model alias is a name you choose that points at any model in the CrossModel catalog. Aliases are configured per API key, so you can give a client a stable model name once and re-point it to a different real model whenever you like — without touching the client.

This is what the name CrossModel is about: you call your own name, and we route it across to the model you picked.

Why use them

Many tools hard-code the model name they send (for example coding assistants that always request a fixed claude-* id). With an alias you can:

  • Point a client at CrossModel once (just the base URL + key), then switch the underlying model from the dashboard.
  • Give every project a stable name like cross-model-basic / cross-model-pro and decide later what each maps to.
  • Remap a hard-coded model name a client insists on sending to whatever you actually want to run.

How resolution works

When a request arrives, the gateway resolves the requested model name against the key's aliases before routing and billing:

  1. Exact match — an alias whose name equals the requested model wins.
  2. Longest prefix match — among prefix aliases whose name is a prefix of the requested model, the longest one wins.
  3. Passthrough — if nothing matches, the original name is used unchanged.

Once resolved, the request is routed and billed exactly as if you had sent the real model id. Aliases never change pricing — you pay for the real model that runs.

Aliases take priority over real model names. If you name an alias the same as a real catalog id, the alias wins — which is exactly what lets you remap a name a client hard-codes.

Exact vs. prefix

  • Exact — cross-model-basic → deepseek/deepseek-chat. Only a request for cross-model-basic matches.
  • Prefix — claude-opus → anthropic/claude-opus-4-8 (prefix). Requests for claude-opus-4-8, claude-opus-4-9, … all match, so a client that bumps its version string keeps working without you editing the alias.

Configuring aliases

Open the API Keys page in the console, click the aliases button on a key, and add rows of alias → target model. The target picker searches your available models. A built-in test box lets you type a model name and see exactly which real model it resolves to.

Alias names must be lowercase and may use letters, digits, and . _ : / - (up to 64 characters). Each key allows up to 20 aliases.

Changes take effect within about a minute (the gateway briefly caches key data).

Listing models

GET /v1/models returns your aliases (each with a root field pointing at the real model id and that model's capabilities) followed by the full catalog. Real model names remain callable alongside your aliases, so tools that read the model list see both.

{
  "id": "cross-model-basic",
  "object": "model",
  "root": "deepseek/deepseek-chat",   // the real model this alias resolves to
  "context_length": 65536,
  "architecture": { "input_modalities": ["text"], "output_modalities": ["text"] }
}

Using an alias

Once configured, just send the alias as the model:

curl https://api.crossmodel.ai/v1/chat/completions \
  -H "Authorization: Bearer $CROSSMODEL_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "cross-model-basic",
    "messages": [{ "role": "user", "content": "Hello" }]
  }'