Get started
Models
Neural Router exposes hundreds of models from every major provider behind one schema. Use them by id, by family, or let the router choose.
Model ids
Every model has a stable id (for example claude-sonnet, gpt-4o, or llama-4-70b). Pass an id as model to pin it, or pass "auto" to let the router select within your objective and policies.
Ensemble models appear in the same list, prefixed nr/moa-. They are not single endpoints: one request to nr/moa-max is answered by several models in parallel and synthesised by one, so it bills as the sum of those hops and its listed price is indicative. See Ensemble (Mixture of Agents).
List models
Query the models available to your workspace, with provider, pricing, context window, and live health:
curl https://api.neuralrouter.ai/v1/models \
-H "Authorization: Bearer $NEURALROUTER_API_KEY"New models appear automatically as providers onboard. You can also compare models publicly on the rankings leaderboard by quality-per-dollar, latency, and uptime.
Restricting models
To constrain which models a workspace may use, enable the model-restriction guardrail under Guardrails. Routing and direct model pins are both limited to the approved set, so the control can't be bypassed per request.
Your own provider keys
Already have provider accounts? Add them under BYOK to route through your own credits and rate limits while keeping one unified API. See Bring your own key.