Use Reasoning Models
Use models that work through a problem before answering: pick one from the catalog, let nrouter/auto detect the ask, or use the Responses endpoint.
Reasoning models spend compute working through a problem — planning, checking, backtracking — before they answer, which suits maths, code, analysis, and multi-step agent work. nRouter gives you three ways to reach that behaviour, and no extra settings of its own: the model you choose is the control.
Name a reasoning model
Reasoning models sit in the same catalog as every other model — the models reference shows filtering the catalog for them (its example picks out the o1 series and DeepSeek R1 by ID). Send the model's ID as a direct request:
curl https://api.nrouter.ai/v1/chat/completions \
-H "Authorization: Bearer $NROUTER_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "deepseek-r1",
"messages": [
{"role": "user", "content": "A bat and a ball cost $1.10 in total. The bat costs $1.00 more than the ball. How much does the ball cost? Show your working."}
]
}'A direct request is never swapped: the model you named serves it, or a fallback you configured yourself answers. Browse the live catalog on the Models page to see which reasoning models are available today.
Let nrouter/auto detect it
A request that asks for reasoning is one of the three signals that mark an nrouter/auto request as heavy — alongside long input and tools carried over many turns — so the router classes it for a model fit to reason. The model that served is named in the x-nr-model header and billed at its list price. See Chat Completions.
Use the pipeline endpoints
For work that reasons across many steps, two endpoints are built for it:
- Responses — documented for autonomous agent loops and complex reasoning pipelines, with multi-step tool execution.
- Messages — direct support for Anthropic's thinking mode, alongside content blocks and tool use.
Related guides
- Use Tool Calling — Functions the model can call while it works
- Get Structured Output — Reasoning you can parse: answers in a strict shape
- Send Image Input — Models that reason over images as well as text