HYBRIDAGENT THINKS IN LATENT SPACE AND ACTS IN TEXT

Summary from the TLDR Dev newsletter

HybridAgent is a research approach for training AI agents to reason internally using numerical representations instead of writing out every intermediate thought, while still producing text to call tools such as Python or web search and deliver answers. In tests, a 14B Qwen3 model trained this way scored 52.4 on GAIA versus 43.7 for a matched text-only agent, with estimated decoding costs falling from $470 to $176 per million questions.