Your Agent Is Overthinking
Extra reasoning often doesn’t change whether a coding agent succeeds. We trained a small model to predict when it will, before the agent starts.
Research
News from Rogue
Research, ideas and what we're building.
Introducing Rogue Constellation
Rogue Constellation puts five agents on every task and returns the solution with the strongest evidence that it works. Our flagship coding agent, available in Rogue.
Research
Learning to DeepThink with LLMs
Parallel thinking with GLM 5.3: higher benchmark scores as effort increases, with submitted patch size staying nearly flat.
Research
Ampliar o que uma pessoa consegue construir
Nossa visão para quem quer transformar uma ideia em software, entender suas escolhas e cuidar do que vem depois.
Company