Hermes: Learning Contextual Reasoning Unlocks Test-Time Scaling

Researchers have introduced Hermes, a method that improves artificial intelligence performance by optimizing how models allocate computational resources during inference. The approach focuses on managing information across multiple context windows, allowing models to perform better by strategically deciding which data to retain and process during test-time scaling.