JetBrains Releases Mellum2, a 12B MoE Open Model for High-Frequency Coding Sub-Tasks

Summary

JetBrains released Mellum2, an open-source (Apache 2.0) mixture-of-experts model with 12 billion total parameters and 2.5 billion active per token, optimized for text and code workloads. The MoE design keeps total capacity high while activating a small subset per token, delivering what JetBrains reports as 2x+ faster inference than similarly sized models with competitive quality across code generation, reasoning, science, and math benchmarks. Weights are available on Hugging Face with a technica