---
id: 20260812-T0-03
title: "推理模型不擅分配算力：跨问题场景下测试时计算效率低下"
title_en: "Reasoning Models Fail to Allocate Test-Time Compute Efficiently Across Questions"
url: https://ai.daily.yangsir.net/daily/20260812-T0-03
issue_date: 2026-08-12
publish_date: 2026-08-11T04:00:00.000Z
category: research
source_name: "arXiv cs.CL (NLP)"
source_url: https://arxiv.org/abs/2608.07968
---

# 推理模型不擅分配算力：跨问题场景下测试时计算效率低下

arXiv新论文指出：推理语言模型在单问题上能有效利用测试时计算提升性能，但在多问题共享算力或延迟约束时，模型无法合理分配计算资源。研究发现，前沿推理模型在跨问题场景下表现显著下降，说明现有推理策略缺乏全局优化能力。

## English Version

**Reasoning Models Fail to Allocate Test-Time Compute Efficiently Across Questions**

New arXiv paper finds reasoning language models struggle to allocate test-time compute when multiple problems share cost or latency constraints. Models show significant degradation in cross-question scenarios, revealing a lack of global optimization in current reasoning strategies.

---

**来源**：[arXiv cs.CL (NLP)](https://arxiv.org/abs/2608.07968)

**详情页**：https://ai.daily.yangsir.net/daily/20260812-T0-03

---

*智语观潮 · Daily — https://ai.daily.yangsir.net/llms.txt*