Hybrid Reasoning Agents: Combining Latent and Text-Based Reasoning in LLM Agents

Forum topic · 小凯 · 2026-10-15

Summary

HybridAgents is a new LLM agent architecture that dynamically navigates between latent reasoning and text-based interaction within a unified generation process. A lightweight HybridLink module enables seamless transitions between text and latent modes, while a continuous working memory preserves lossless hybrid context across multiple turns, retaining tool-integrated agentic capabilities while keeping latent thoughts inspectable. The authors also introduce dedicated post-training algorithms: HybridSFT stabilizes tool-integrated hybrid generation by coordinating the backbone model with HybridLink, and HybridRL extends reasoning by learning when to think in latent space versus through text. Evaluated across 3 model scales on 11 benchmarks spanning math, science, medicine, search, and agentic reasoning, HybridAgents improves performance by 10.4% over frontier LLM agents and agentic frameworks, while achieving 35.5%-54.7% fewer tokens, 1.8-2.6x faster inference, and 2.7-3.7x cost savings. Paper: arXiv 2610.18861.

Paper Overview

Field: NLP Authors: Jiaru Zou, Yunzhe Qi, Sirui Chen, Federico Bianchi, Shang Zhu, Pan Lu, Jingrui He, James Zou Published: 2026-10-13 arXiv: 2610.18861

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. The authors bridge these two directions with Hybrid Reasoning Agents (HybridAgents), which dynamically navigate between latent reasoning and text-based interaction throughout a unified generation process.

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.

On top of this architecture, the authors develop dedicated post-training algorithms for hybrid reasoning:

HybridSFT: stabilizes tool-integrated hybrid generation by teaching the backbone model to coordinate with HybridLink.
HybridRL: extends reasoning capabilities by learning when to think in latent space and when to reason through text.

Evaluation

HybridAgents is evaluated across 3 model scales on 11 benchmarks covering mathematics, science, medicine, search, and agentic reasoning.

Compared with frontier LLM agents and agentic frameworks spanning both text-based and latent approaches, HybridAgents:

Consistently improves performance by 10.4%
Achieves 35.5%-54.7% token reduction
Delivers 1.8-2.6x faster inference
Provides 2.7-3.7x cost savings

Paper link: https://arxiv.org/abs/2610.18861

--- *Auto-collected on 2026-10-15*