Choose the workload
Select a model and a traffic slice with known costs and expected output.
A measured path to lower inference cost.
Evaluate the models, costs, and operational behavior your team depends on before moving more traffic.
Enterprise / Start a conversation
Make your next evaluation count.
Share your model mix, expected volume, and requirements. Your inquiry goes into the router review queue.
- Model and protocol compatibility
- Prices and request-level billing evidence
- Latency, routing, and operational requirements
- Data handling and supplier coverage
Submitting an inquiry does not create a paid agreement or reserve capacity.
Start small. Expand with evidence.
A representative workload gives engineering and finance a shared basis for deciding what to move next.
Connect your client
Use your router API key with Chat Completions, Responses, or Messages.
Review the result
Compare settled cost, token use, request status, and observed latency.
Know where your request goes.
Metering records
Request records include model, token counts, charges, status, and timing. The metering tables do not contain prompt or response bodies. Saved experiments contain the instructions you configure. Optional exact caching stores eligible response text with a 15-minute reuse window; account settings and explicit experiment overrides control it.
Read privacy informationServing providers
The selected provider processes request content. Supplier identities are private in the public catalog; contract and retention review is a separate step.
Review processing boundariesVerified coverage
Zero data retention is not currently verified for these routes. A request requiring ZDR fails when no verified provider is eligible.
Read the routing controls