Topic
LLM agents
Latest news
- 19 OctHybridReasoning Agents blend latent thinking with text-based actions
Researchers propose HybridAgent, an architecture that lets agents alternate between latent-space reasoning and text actions while keeping tool access and inspectability. Evaluated across scales and 11 benchmarks, HybridAgent reportedly improves performance by 10.4% and reduces token usage by 35.5%–54.7% compared with prior agent methods.
Research · 1 source
Articles
No articles about LLM agents yet.