Introducing Mellum2: A 12B Mixture-of-Experts Model by JetBrains
Source
huggingface.co
Date
Context
Mellum2 is a text-and-code model designed for frequent, bounded tasks inside larger AI systems, such as routing requests or summarizing retrieved material. JetBrains' release article describes a mixture-of-expertsA model built from many specialist sub-networks where only a few activate per token, giving big-model capability at small-model running cost.Full definition → architecture with 12 billion parameters in total and 2.5 billion active for each generated token. Only a subset of the model's learned parameters participates in each step.
That design targets the repeated intermediate calls that can accumulate latency in an AI agentAn AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.Full definition → workflow. The model is released under Apache 2.0 and supports self-hosted deployment, giving teams an option for running these text-and-code tasks on their own infrastructure.
Terms in this piece · Glossary
mixture-of-experts — A model built from many specialist sub-networks where only a few activate per token, giving big-model capability at small-model running cost.
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
RAG — Retrieval-augmented generation — fetching relevant documents first and pasting them into the model's context so it answers from your data instead of memory.
subagent — A separate agent spawned by another to do one scoped piece of work in its own context, returning only the result.
