Decoded
Feed

Hybrid Reasoning Agents

arXiv:2610.18861CLAILG13 oct 20261 min readHN ×7PDF

An AI agent that thinks silently in vectors but still writes text to use tools.

10.4%

Average performance improvement

Gains while using 35.5%–54.7% fewer tokens than text agents

How it works

  1. Two worlds apart

    Text agents can use tools but spell out every thought. Latent reasoners think faster in vectors but cannot call tools.

  2. Switch modes mid-generation

    A small adapter, HybridLink, feeds hidden states back as inputs. Special tokens open and close silent latent spans.

  3. Better, cheaper agents

    Qwen3 models at 4B to 14B beat text and latent baselines on 11 benchmarks with faster inference.

<latent></latent>actobservationsnext turnQuestioninputText modeHybridLinklatent loopTool callsKV workingmemoryFinal answer<latent></latent>actobservationsnext turnQuestioninputText modeHybridLinklatent loopTool callsKV workingmemoryFinal answer
From the paper