JetBrains has released Mellum2.1, an open model designed for coding agents that work inside software repositories. The company says it can explore a codebase, edit files and check its changes. The model is available on Hugging Face.
Mellum2.1 keeps the architecture of Mellum2, the model JetBrains open-sourced in June. It is a 12 billion parameter mixture-of-experts model with 2.5 billion active parameters, released under the Apache 2.0 license. JetBrains says the update focused on training after the model’s initial training stage.
Training adds repository tasks
JetBrains says reinforcement learning became the main part of training for Mellum2.1. It launched millions of sandboxed runs across thousands of environments to train the model on tasks that involve working with tools and software repositories.
The company also added tasks in math, competitive programming, science, tool use and software engineering. It says it filtered the training sources for problems such as broken tests and answers that could not be verified.
JetBrains reports coding and speed gains
In evaluations using the same setup for Mellum2.1, Mellum2, Qwen3.5-9B and Gemma 4 E4B, JetBrains says the largest improvement over Mellum2 was in agentic coding. It also reports gains in coding, competitive programming, math, tool calling and general knowledge.
JetBrains says Mellum2.1 was the fastest of those models under heavy load, serving almost twice as many tokens as Qwen3.5-9B. It says multi-token prediction makes single requests about 1.6 times faster.
JetBrains says developers can run Mellum2.1 locally or on their own infrastructure. GGUF builds for llama.cpp, Ollama and LM Studio, along with a multi-token prediction head for speculative decoding in vLLM, are coming soon.
