Changes from previous version
Expands beyond code completion to natural language and broader software-engineering workflows; MoE architecture targets higher throughput than the 4B dense Mellum line.
Release Summary
12B-parameter Mixture-of-Experts focal model (2.5B active per token) for routing, RAG, sub-agents, and private deployments. Ships open from day one with base, instruct, and thinking checkpoints under Apache 2.0.
Timeline
June 1, 2026
Mellum2 open-sourced
JetBrains releases Mellum2 on Hugging Face with base, instruct, and thinking variants plus a technical report.