Models and groups

Discover what your key can call, and why the same model name can behave differently across groups.

Model names are passed through to the upstream provider, so use the provider’s identifier: gpt-4o-mini, claude-sonnet-4-5, gemini-2.5-flash, dall-e-3, and so on.

Discover what you can call

curl https://router.triw.io/v1/models \
  -H "Authorization: Bearer $TRIW_API_KEY"

The response is a standard OpenAI model list, filtered to the models your key’s group may use. Treat this endpoint as authoritative — it reflects live availability, not a static page.

Groups in one paragraph

A group is the operator-defined profile that decides which upstream channels a key may use and what it pays. A model that exists globally but not in your group returns:

{"error":{"message":"model not found","type":"new_api_error","code":"model_not_found"}}

Fix: create a key in a group that includes the model, then re-check /v1/models with that key.

Model families Triw relays

Category	Typical identifiers	Endpoint
Chat / reasoning	gpt-*, claude-*, gemini-*, deepseek-*, qwen-*	/v1/chat/completions, /v1/messages, /v1beta/...
Embeddings	text-embedding-*, embedding-*	/v1/embeddings
Image generation	dall-e-3, dall-e-2, gpt-image-*, flux-*	/v1/images/generations
Audio	whisper-1, tts-1 and provider equivalents	/v1/audio/*
Rerank	rerank-*	/v1/rerank

Which identifiers actually resolve depends on the channels the operator has connected. /v1/models is the answer for your key.

Choosing a model name

• Use the exact string from /v1/models. Typos surface as model_not_found, not as a silent fallback.
• For OpenAI-compatible clients, model names with provider prefixes are not needed — send the bare identifier.
• For Anthropic clients, send claude-* ids; Triw translates the request to whichever channel serves that id.