New: Model Router Benchmarks Compare 7 routers side by side on quality, speed,…

A model router chooses which AI model should handle a request, but making that choice adds its own costs. OpenRouter's benchmark methodology explains that switching models makes requests more expensive because the input cache is rebuilt, while the more processing goes into routing decisions, the more latency is introduced. Its comparison shows individual models alongside routers as baselines, so readers can examine whether routing earns back that overhead.

The combined Router Index also expresses a preference: its default weights are quality at 60%, time per task at 20% and cost at 20%, and readers can adjust them with a slider. A router's overall position therefore reflects both measured performance and how those competing objectives are valued.