JetBrains has introduced Mellum2, a 12B Mixture-of-Experts model optimized for efficient text-and-code tasks, offering faster inference and lower latency for software engineering workloads. The open model excels in routing, retrieval-augmented generation (RAG) pipelines, and sub-agent tasks while being deployable in private environments. Designed for specialized use rather than replacing larger models, Mellum2 aims to enhance AI system efficiency and cost-effectiveness.

JetBrains has released Mellum2, an open 12-billion-parameter Mixture-of-Experts model optimized for low-latency text and code tasks. The company says it performs competitively with similarly sized open models while delivering more than twice the inference speed, making it suited for high-throughput production workloads. Designed as a "focal" model for software engineering, Mellum2 targets uses such as routing and orchestration, RAG pipelines, agent subtasks, and private self-hosted deployment.