Menlo Models

At Menlo, we have focused on creating a series of models that are optimized for all sorts of tasks, including web search, deep research, robotic control, and using MCPs. Our latest model, Jan-Nano-Gguf, is available in Jan right now providing excellent results on tasks that use MCPs.

You can have a look at all of our models, and download them from the HuggingFace Menlo Models page (opens in a new tab).

Jan-Nano-Gguf (Available in Jan right now 🚀)

Jan-Nano-Gguf is a 4-billion parameter model that is optimized for deep research tasks. It has been trained on a variety of datasets and is designed to be used with the Model Context Protocol (MCP) servers.

ReZero

ReZero (Retry-Zero) is a reinforcement learning framework that improves RAG systems by rewarding LLMs for retrying failed queries. Traditional RAG approaches struggle when initial searches fail, but ReZero encourages persistence and alternative strategies. This increases accuracy from 25% to 46.88% in complex information-seeking tasks.