Hybrid Reasoning Agents

<p>This paper introduces <strong>HybridAgents</strong>, a novel framework that blends continuous latent reasoning with explicit text interactions within a unified generation framework. By utilizing a lightweight adapter called <strong>HybridLink</strong>, the system seamlessly transitions between internal thinking modes and external tool usage without losing crucial context. The model leverages continuous working memory to maintain <strong>multi-turn hybrid generation</strong>, effectively eliminating the need for traditional token-space reconstruction across different conversational steps. Furthermore, specialized post-training protocols involving <strong>Hybrid Supervised Fine-Tuning</strong> and <strong>Hybrid Reinforcement Learning</strong> stabilize generation and scale reasoning capabilities. Extensive empirical evaluations demonstrate that this architecture significantly outperforms standard language models, achieving superior accuracy while concurrently reducing <strong>inference latency</strong>, <strong>token consumption</strong>, and <strong>computational overhead</strong>.</p>