Model router in Foundry Models: start in one line, route intelligently, evaluate cost and quality
Model router in Foundry Models adds Auto (preview): set the model to auto and the router picks a model for each request with nothing to deploy. The routing pool now has 45 models from OpenAI, Anthropic, Microsoft AI, xAI, DeepSeek, Meta and Fireworks. Also in preview: per-request routing controls, adaptive session affinity that keeps a conversation on a model while its prompt cache is warm, your own PTU deployments serving routed requests first, and automatic model selection inside the now generally available AI Gateway.
- status
- Public preview
Most agent calls are routine and don't need a frontier model, so routing is one of the easiest cost levers available. The real value is that you can now check that it works. Run an evaluation of the router against a single-model baseline on your own data, and the deployment's Monitor tab shows estimated cost next to what the same traffic would have cost on one model. Pin the model subset to the models you've approved; routing quality is Microsoft's estimate, not a guarantee for your workload.
Official sources
- techcommunity.microsoft.comModel router in Foundry Models: start in one line, route intelligently, evaluate cost and quality