RouterPlus

RouterPlus is an inference gateway that picks the right model and provider for each workload a team is running, aimed at engineering and ML teams who are juggling multiple LLMs across production. It routes on your own evals and production traffic, gives you every model worth running behind one key, and lines up the capacity you will need next. Instead of choosing a single frontier model and hoping it generalizes, teams can score candidates against their own quality bar and send different tasks (support triage, extraction, code review, agents) to whichever model clears it at the best price. The product covers both closed and open-weight models behind one API, with data compliance rules enforced on every request, regional routing, and production-grade SLAs. There are two billing paths: a marketplace that uses prepaid credits at listed model rates, and a bring-your-own-key mode where the underlying provider bills you directly and RouterPlus charges no inference fee. Existing OpenAI or Anthropic-compatible clients can connect by swapping the base URL and key, and an Optimize feature compares candidate setups against a baseline on held-out eval cases before anything is deployed. Reserved capacity is available for teams planning migrations between closed and open models or onboarding new providers. The launch matters now because model churn has become a weekly operational problem for AI teams, and single-vendor commitments are getting expensive to maintain as better open-weight options keep landing. RouterPlus is part of the YC F26 batch and the launch was posted by Alok Bishoyi, an IIT Bombay alum who is building the company.