When Routing Collapses: On the Degenerate Convergence of LLM Routers
Guannan Lai, Han-Jia Ye · 2026
Read When Routing Collapses: On the Degenerate Convergence of LLM Routers by Guannan Lai, Han-Jia Ye (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-03478.txt.
Description
Routing a question to a small cheap model or a large expensive one is supposed to save money. Guannan Lai and Han-Jia Ye document a failure of that idea in a 2026 paper. As budgets grow, routers default to the priciest model even when a cheaper one would do. They name this failure routing collapse, and it quietly wastes computation.
The authors trace the problem to a mismatch between training and decisions. Routers learn to predict scalar scores, but routing is a discrete comparison, so small scoring errors flip the ranking. Their EquiRouter learns model rankings directly instead. On RouterBench it reaches GPT-4-level quality at roughly 17 percent lower cost than the strongest earlier router.
Catalog record
| CREATOR | Guannan Lai, Han-Jia Ye |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 62 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-03478