Model identifiers & vendor prefixes
How ToRouter names models — OpenAI-style IDs by default, vendor-prefixed names for OpenRouter channels, plus the -thinking suffix.
ToRouter accepts OpenAI-style model IDs in every request. The gateway maps the ID to whichever upstream channel your group is bound to.
Canonical IDs
Use bare IDs without a vendor prefix:
gpt-5
gpt-5-mini
o3
claude-opus-4-7
claude-sonnet-4-7
gemini-3-flash-preview
gemini-2.5-pro
text-embedding-3-largeThese work uniformly across OpenAI, Anthropic and Gemini protocol surfaces — the gateway translates as needed.
Vendor prefixes (OpenRouter channels)
When your group is bound to an OpenRouter account, the OpenRouter-style vendor/model form is also accepted and passed through unchanged:
anthropic/claude-opus-4-7
google/gemini-2.5-pro
openai/gpt-5
deepseek/deepseek-r1This is the only case where vendor prefixes are meaningful — on OpenAI/Anthropic/Gemini channels, prefer the bare ID.
The -thinking suffix
For models that support extended reasoning, append -thinking to enable it:
claude-opus-4-7-thinking
gpt-5-thinkingToRouter strips the suffix before forwarding and injects the upstream's reasoning flag. Token usage and pricing reflect reasoning tokens.
The catalog at Models · Catalog shows which models support -thinking and which vendor prefixes your group accepts.
Listing models
curl https://portal.torouter.ai/v1/models \
-H "Authorization: Bearer sk-***"Returns every model your subscribed group is allowed to call.