Berkeley Study Finds Model Routing Cuts Costs Up to 25x
UC Berkeley and partner labs released a large-scale model routing study spanning 452 candidate models and policies across 25 benchmarks, totaling 374,401 scored runs. Routers deliver their value mainly by shifting work to cheaper models, averaging 8.82x more quality per unit cost with under 0.5 points lost; on coding tasks routing reached 25.5x the cost efficiency of always calling Opus 4.6 at a 0.44-point loss. Yet learned routers gained only 0.3 points over the best single model, far below the 5.4-point theoretical ceiling.
Related Entities
Full Report
Rohan Paul (@rohanpaul_ai) · 2026-11-10 17
Great new study on model routing.
From University of California, Berkeley + other labs
Finds that model routers pay off significantly by saving money, averaging 8.82X more quality per unit cost while losing under 0.5 points.
Routers pick which model handles each input, hoping to match the best model for less or beat it. Most routing research has studied only language models.
Route Anything ran 452 candidate models and policies on the same inputs across 25 benchmarks, from text QA to robot control. They trained routers on these 374,401 scored runs and measured both quality and cost.
A perfect per-input router would beat the best single model by 5.4 points, but learned routers gained only 0.3 across 16 benchmarks. On coding tasks, routing reached 25.5× the dollar efficiency of always calling Opus 4.6 at a 0.44-point loss.
Use routers mainly to move work to cheaper models when quality is close, and compare candidates on the same inputs before trusting gains.
