dorsal/arxiv
View SchemaBreaking Model Lock-in: Cost-Efficient Zero-Shot LLM Routing via a Universal Latent Space
| Authors | Cheng Yan, Wuyang Zhang, Zhiyuan Ning, Fan Xu, Ziyang Tao, Lu Zhang, Bing Yin, Yanyong Zhang |
|---|---|
| Categories | |
| ArXiv ID | 2601.06220vv1 |
| URL | https://arxiv.org/abs/2601.06220 |
| License | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ |
Abstract
The rapid proliferation of Large Language Models (LLMs) has led to a fragmented and inefficient ecosystem, a state of ``model lock-in'' where seamlessly integrating novel models remains a significant bottleneck. Current routing frameworks require exhaustive, costly retraining, hindering scalability and adaptability. We introduce ZeroRouter, a new paradigm for LLM routing that breaks this lock-in. Our approach is founded on a universal latent space, a model-agnostic representation of query difficulty that fundamentally decouples the characterization of a query from the profiling of a model. This allows for zero-shot onboarding of new models without full-scale retraining. ZeroRouter features a context-aware predictor that maps queries to this universal space and a dual-mode optimizer that balances accuracy, cost, and latency. Our framework consistently outperforms all baselines, delivering higher accuracy at lower cost and latency.
{
"annotation_id": "e6747461-5ee4-422d-af55-018102f4ba9a",
"date_created": "2026-02-17T05:53:08.675000Z",
"date_modified": "2026-02-17T05:53:08.675000Z",
"file_hash": "cb5b08b998f5609f81591eab66c1a1b93ae13c55cc4c5853f4e7d523348564d5",
"private": false,
"record": {
"abstract": "The rapid proliferation of Large Language Models (LLMs) has led to a fragmented and inefficient ecosystem, a state of ``model lock-in\u0027\u0027 where seamlessly integrating novel models remains a significant bottleneck. Current routing frameworks require exhaustive, costly retraining, hindering scalability and adaptability. We introduce ZeroRouter, a new paradigm for LLM routing that breaks this lock-in. Our approach is founded on a universal latent space, a model-agnostic representation of query difficulty that fundamentally decouples the characterization of a query from the profiling of a model. This allows for zero-shot onboarding of new models without full-scale retraining. ZeroRouter features a context-aware predictor that maps queries to this universal space and a dual-mode optimizer that balances accuracy, cost, and latency. Our framework consistently outperforms all baselines, delivering higher accuracy at lower cost and latency.",
"arxiv_id": "2601.06220",
"authors": [
"Cheng Yan",
"Wuyang Zhang",
"Zhiyuan Ning",
"Fan Xu",
"Ziyang Tao",
"Lu Zhang",
"Bing Yin",
"Yanyong Zhang"
],
"categories": [
"cs.LG",
"cs.AI"
],
"license": "http://arxiv.org/licenses/nonexclusive-distrib/1.0/",
"title": "Breaking Model Lock-in: Cost-Efficient Zero-Shot LLM Routing via a Universal Latent Space",
"url": "https://arxiv.org/abs/2601.06220",
"version": "v1"
},
"schema_id": "dorsal/arxiv",
"source": {
"execution_id": "36c9f546-f379-483d-af8b-61d79aa0789c",
"id": "arXiv Dataset",
"type": "Model",
"variant": "snapshot-2026-01-17",
"version": "0.1.0"
},
"user_id": 1000002
}