Hybrid Reasoning Agents: 35–55% fewer tokens and up to 2.6× faster inference

Hybrid Reasoning Agents (HybridAgents) bridge latent reasoning and text-based interaction by dynamically navigating between the two modes through a lightweight HybridLink. Evaluated across 11 benchmarks, the architecture 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.