Select-then-Solve: Paradigm Routing as Inference-Time Optimization for LLM Agents
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , , , , , , , , , , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2026
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866908947388039168 |
|---|---|
| author | Zhou, Heng Tan, Zelin Zhang, Zhemeng Fan, Yutao Lin, Yibing Kang, Li Song, Xiufeng Li, Rui Huang, Songtao Yu, Ao Fan, Yuchen Chen, Yanxu Xu, Kaixin Liu, Xiaohong Qin, Yiran Torr, Philip Zhang, Chen Yin, Zhenfei |
| author_facet | Zhou, Heng Tan, Zelin Zhang, Zhemeng Fan, Yutao Lin, Yibing Kang, Li Song, Xiufeng Li, Rui Huang, Songtao Yu, Ao Fan, Yuchen Chen, Yanxu Xu, Kaixin Liu, Xiaohong Qin, Yiran Torr, Philip Zhang, Chen Yin, Zhenfei |
| contents | When an LLM-based agent improves on a task, is the gain from the model itself or from the reasoning paradigm wrapped around it? We study this question by comparing six inference-time paradigms, namely Direct, CoT, ReAct, Plan-Execute, Reflection, and ReCode, across four frontier LLMs and ten benchmarks, yielding roughly 18,000 runs. We find that reasoning structure helps dramatically on some tasks but hurts on others: ReAct improves over Direct by 44pp on GAIA, while CoT degrades performance by 15pp on HumanEval. No single paradigm dominates, and oracle per-task selection beats the best fixed paradigm by 17.1pp on average. Motivated by this complementarity, we propose a select-then-solve approach: before answering each task, a lightweight embedding-based router selects the most suitable paradigm. Across four models, the router improves average accuracy from 47.6% to 53.1%, outperforming the best fixed paradigm at 50.3% by 2.8pp and recovering up to 37% of the oracle gap. In contrast, zero-shot self-routing only works for GPT-5 at 67.1% and fails for weaker models, all trailing the learned router. Our results argue that reasoning paradigm selection should be a per-task decision made by a learned router, not a fixed architectural choice. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2604_06753 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Select-then-Solve: Paradigm Routing as Inference-Time Optimization for LLM Agents Zhou, Heng Tan, Zelin Zhang, Zhemeng Fan, Yutao Lin, Yibing Kang, Li Song, Xiufeng Li, Rui Huang, Songtao Yu, Ao Fan, Yuchen Chen, Yanxu Xu, Kaixin Liu, Xiaohong Qin, Yiran Torr, Philip Zhang, Chen Yin, Zhenfei Computation and Language When an LLM-based agent improves on a task, is the gain from the model itself or from the reasoning paradigm wrapped around it? We study this question by comparing six inference-time paradigms, namely Direct, CoT, ReAct, Plan-Execute, Reflection, and ReCode, across four frontier LLMs and ten benchmarks, yielding roughly 18,000 runs. We find that reasoning structure helps dramatically on some tasks but hurts on others: ReAct improves over Direct by 44pp on GAIA, while CoT degrades performance by 15pp on HumanEval. No single paradigm dominates, and oracle per-task selection beats the best fixed paradigm by 17.1pp on average. Motivated by this complementarity, we propose a select-then-solve approach: before answering each task, a lightweight embedding-based router selects the most suitable paradigm. Across four models, the router improves average accuracy from 47.6% to 53.1%, outperforming the best fixed paradigm at 50.3% by 2.8pp and recovering up to 37% of the oracle gap. In contrast, zero-shot self-routing only works for GPT-5 at 67.1% and fails for weaker models, all trailing the learned router. Our results argue that reasoning paradigm selection should be a per-task decision made by a learned router, not a fixed architectural choice. |
| title | Select-then-Solve: Paradigm Routing as Inference-Time Optimization for LLM Agents |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2604.06753 |