DeepMind optimizes model routing using the Pandora's Box problem

dair_ai · x · 2026-08-21

Google DeepMind published a significant paper on model routing. Existing routers assume value estimation is free, but estimation is often the most expensive part of the pipeline.

The research formalizes model routing as the "Pandora's Box" problem—optimal search when inspection is expensive. Under a Gaussian signal model, the policies have closed-form solutions, indicating exactly when refining an estimate is worth the cost for each specialist and input.

Across multi-LLM benchmarks, retrieval-augmented specialists, and LLMs with variable inference-time reasoning, Pandora's Router matches the quality of exhaustive estimation while calling the expensive estimator far less often. When competing estimates are noisy, value-of-information reasoning increases the strategic specialist's utility.

Original post →

More from Infra

Infra channel →