RouterPlus is built
for your workload.
Routing that knows
your workload.
Routed on your own evals and production traffic —, not on global benchmarks.
Read the benchmarkYour future inference
needs, lined up.
- Reserved capacity
- Your traffic
- 01 Reserve steps upMoving from closed to open source
- 02 Reserve steps upUpgrading to a better open model
- 03 Reserve steps upAdding your next provider
Talk to us about
your inference needs.
Questions teams
ask first.
Still have questions?
Talk it through with the RouterPlus team. 30 minutes, tailored to your workload.
Book a callRouterPlus is for teams building with multiple AI models or looking for a better fit for their workload. Use one API key across providers, connect your own provider accounts, or compare models on your own evals.
Optimize compares candidates with your baseline on held-out eval cases. It only marks a winner when the candidate passes its quality and cost checks. You choose whether to deploy it; earlier deployed versions remain available.
Start with an API key and a model from the catalog. Existing OpenAI- or Anthropic-compatible clients can connect by changing the base URL and key. Finding a better setup depends on your evals, workload, and search budget.
With bring-your-own-key routing, supported provider accounts continue billing you directly. Connect your existing credentials instead of moving that spend to RouterPlus credits. Talk to us about any reserved-capacity requirements.
The gateway does not store prompts or completions; it records usage metadata. The provider serving your request applies its own data policy. Optimize separately stores your eval cases and candidate answers until you delete the search.
Marketplace calls use prepaid credits at the listed model rates, with the cost included in each billed response. For BYOK inference, your provider bills you directly and RouterPlus charges no inference fee. Check the model catalog and docs for current details.