Google Cloud adds a router across AI models

Google Cloud’s API Gateway now offers model routing in public preview, letting developers dynamically route traffic to models including Gemini, Claude, and open-weight models like OSS-GPT through a single unified API rather than hardcoding a specific provider’s endpoint into their application. The feature is aimed at teams running multi-model setups who currently maintain separate integration code, retry logic, and failover handling for each provider they use, the same problem that’s driven Chinese open models onto OpenRouter as developers hunt for the best price-to-capability ratio across providers.

What it means for operators

Model routing has mostly lived as a third-party layer, tools built specifically to arbitrage between providers on cost, latency, or capability. Google folding it into its own API Gateway signals the cloud platforms themselves now see multi-model routing as core infrastructure rather than a niche add-on. If you’re running production workloads across more than one model provider, it’s worth comparing this against whatever routing setup you already have, since a first-party option inside your existing cloud stack changes the calculus on maintaining custom routing code yourself.