Chain-of-Route
The core routing algorithm of the MultiTurnRoute platform. It uses a learned gating mechanism that combines recent and recurrent dialogue history to predict the response quality of each model, and estimates cost based on reusable prefix length.
Timeline (last 90 days)
The Chain-of-Route (CoR) algorithm uses a learned gating mechanism that examines both recent conversation history and recurring conversation history to predict the reply quality of each model.
CoR's cost component is estimated based on each model's reusable prefix length rather than simply accumulating token counts.
On this benchmark, when reaching the quality level of the strongest single model, CoR reduces normalized cost by up to 20.3% relative to the strongest routing baseline, while peak reply quality is on par or improved.
All Facts (3)
The Chain-of-Route (CoR) algorithm uses a learned gating mechanism that examines both recent conversation history and recurring conversation history to predict the reply quality of each model.
50%UnverifiedCoR's cost component is estimated based on each model's reusable prefix length rather than simply accumulating token counts.
50%UnverifiedOn this benchmark, when reaching the quality level of the strongest single model, CoR reduces normalized cost by up to 20.3% relative to the strongest routing baseline, while peak reply quality is on par or improved.
50%