Choosing a model

Models are addressed by their public ID, exactly as shown in the Model catalog — set that ID as model in your request. There's no separate lookup step: the ID you copy from the catalog is the ID you send. There is no automatic fallback or load balancing — the ID you send is the model you get, unless you deliberately send an Auto Mode ID and ask to be routed.

Model ID format

Most models are addressed by their bare name:

gemini-2.5-flash
nova-pro
claude-haiku-4-5-20251001
kimi-k2.6
MiniMax-M2.5
gpt-5.5

A model offered at more than one context window is listed once per window, with the window appended to the ID, so you pick the exact one you want. Hypothetically, a model called some-model offered at both 200k and 1M tokens would appear in the catalog as two separate entries, some-model-200k and some-model-1m — check the live catalog for which IDs actually carry a window suffix today.

Switching models

Because addressing is just a string, switching models is a one-line change:

model: 'gemini-2.5-flash'
// … or
model: 'kimi-k2.6'

Picking the right one

• Cheapest general-purpose: gemini-2.5-flash
• Long context: models with large context windows — check the contextLength field in the catalog
• Anthropic Claude: claude-* (see Integrations → Anthropic)
• Don't want to choose: send an Auto Mode ID (auto-cheap, auto, auto-quality) and a router picks a model inside that price band per request
• Reasoning models: gpt-5.* and similar — accept reasoning_effort or thinking depending on the model, see Model notes

Browse everything with live pricing and context windows in the Model catalog, and check Model notes for per-model quirks.