Research map / Large language models / Language, safety and interpretability
Breaking Model Lock-in: Cost-Efficient Zero-Shot LLM Routing via a Universal Latent Space
Cheng Yan, Wuyang Zhang, Zhiyuan Ning, Fan Xu, Ziyang Tao, Lu Zhang, Bing Yin, Yanyong Zhang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| select anchor queries by D-optimality instead of random sampling (ablation) | ZeroRouter · New-model onboarding, 200 anchors, Max-Acc | reward: 0.27 -> 0.39 | Improved |
| select anchors by difficulty only (Diff-based) instead of D-optimality (ablation) | ZeroRouter · New-model onboarding, 200 anchors, Max-Acc | reward: 0.39 -> 0.24 | Hurt |
| select anchors by discrimination only (Disc-based) instead of D-optimality (ablation) | ZeroRouter · New-model onboarding, 200 anchors, Max-Acc | reward: 0.39 -> 0.28 | Hurt |
| use task-aware difficulty (alpha^T b) for anchor selection instead of D-optimality (ablation) | ZeroRouter · New-model onboarding, 200 anchors, Max-Acc | reward: 0.39 -> 0.32 | Hurt |
| use D-optimality anchors on mean reward vs random sampling (ablation) | ZeroRouter · New-model onboarding, 200 anchors, mean | reward: -0.25 -> -0.11 | Improved |
| use ZeroRouter instead of GraphRouter, FORC, RouteLLM, Model-SAT on smaller LLMs | 5 smaller LLMs router · ID and OOD, three policies | mean reward: -0.14 -> 0.00 | Improved |
| use ZeroRouter instead of GraphRouter on larger LLMs | 5 larger LLMs router · ID and OOD, three policies | mean reward: -0.07 -> 0.05 | Improved |
| use ZeroRouter instead of strongest baseline on OOD Max-Acc with larger LLMs | 5 larger LLMs router · OOD, Max-Acc | reward: 0.62 -> 0.68 | Improved |
| use ZeroRouter instead of Model-SAT on OOD Max-Acc with smaller LLMs | 5 smaller LLMs router · OOD, Max-Acc | reward: 0.57 -> 0.59 | No clear effect |
Closest papers by meaning
- Universal Model Routing for Efficient LLM Inference (ICLR 2026 · similarity 0.92)
- Lookahead Routing for Large Language Models (NeurIPS 2025 · similarity 0.91)
- GraphRouter: A Graph-based Router for LLM Selections (ICLR 2025 · similarity 0.90)
- Dr.LLM: Dynamic Layer Routing in LLMs (ICLR 2026 · similarity 0.90)
- Confidence-Guided Stepwise Model Routing for Cost-Efficient Reasoning (AAAI 2026 · similarity 0.90)
- RouteLLM: Learning to Route LLMs from Preference Data (ICLR 2025 · similarity 0.90)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
