Hybrid Reasoning Agents: LLM agents that think in latent space, act through text

Computer Science > Computation and Language arXiv:2610.18861 (cs) [Submitted on 13 Oct 2026] Title: Hybrid Reasoning Agents Authors: Jiaru Zou, Yunzhe Qi, Sirui Chen, Federico Bianchi, Shang Zhu, Pan Lu, Jingrui He, James Zou.

计算机科学 > 计算与语言 arXiv:2610.18861 (cs) [提交于 2026 年 10 月 13 日] 标题：混合推理智能体 作者：Jiaru Zou, Yunzhe Qi, Sirui Chen, Federico Bianchi, Shang Zhu, Pan Lu, Jingrui He, James Zou。

Abstract: How can we build an agent that thinks beyond human language, yet still acts through it in the external environment? Modern LLM agents still reason and interact through text, while latent reasoning offers an efficient internal medium but lacks native tool access and reasoning transparency. We bridge these two directions with Hybrid Reasoning Agents (HybridAgents), which dynamically navigate between latent reasoning and text-based interaction throughout a unified generation process.

摘要：我们如何构建一个能够超越人类语言进行思考，同时又能在外部环境中通过语言进行交互的智能体？现代大语言模型（LLM）智能体仍然通过文本进行推理和交互，而潜在推理虽然提供了一种高效的内部媒介，却缺乏原生的工具访问能力和推理透明度。我们通过混合推理智能体（HybridAgents）弥合了这两个方向，它在统一的生成过程中动态地在潜在推理和基于文本的交互之间进行切换。

Architecturally, a lightweight HybridLink enables seamless transitions between text and latent modes, while the agent’s continuous working memory preserves lossless hybrid context across multiple turns. These designs preserve tool-integrated agentic capabilities while making latent thoughts directly inspectable.

在架构上，轻量级的 HybridLink 实现了文本模式与潜在模式之间的无缝转换，同时智能体的连续工作记忆在多轮对话中保留了无损的混合上下文。这些设计在保留工具集成智能体能力的同时，使潜在思维可以直接被检查。

Building on this architecture, we further develop dedicated post-training algorithms for hybrid reasoning: HybridSFT stabilizes tool-integrated hybrid generation by teaching the backbone to coordinate with HybridLink, then HybridRL scales reasoning by learning when to think in latent space and when to act through text.

基于此架构，我们进一步开发了专门用于混合推理的后训练算法：HybridSFT 通过教导骨干模型与 HybridLink 协作，稳定了工具集成的混合生成；随后，HybridRL 通过学习何时在潜在空间思考以及何时通过文本行动，扩展了推理能力。

We evaluate HybridAgent at 3 model scales across 11 benchmarks covering mathematics, science, medicine, search, and agentic reasoning. Compared with frontier LLM Agents and agentic frameworks spanning both text-based and latent methods, HybridAgent consistently improves performance by 10.4%, while delivering 35.5%-54.7% token reduction, 1.8x-2.6x faster inference, and 2.7x-3.7x cost savings.

我们在涵盖数学、科学、医学、搜索和智能体推理的 11 个基准测试中，对 3 种模型规模的 HybridAgent 进行了评估。与涵盖基于文本和潜在方法的尖端 LLM 智能体及智能体框架相比，HybridAgent 的性能持续提升了 10.4%，同时实现了 35.5%-54.7% 的 Token 减少、1.8 倍-2.6 倍的推理加速以及 2.7 倍-3.7 倍的成本节约。