JetBrains Open-Sources Mellum2: A Fast, Specialized AI Model for Production Workflows

JetBrains has made a significant contribution to the AI community by open-sourcing Mellum2, a 12-billion-parameter model specifically engineered to tackle the critical performance bottlenecks in production AI systems: latency, throughput, and operational cost. The model, which has been developed from the ground up, is now available under the permissive Apache 2.0 license, encouraging broad adoption and further development by the community. Mellum2 represents an evolution from its initial focus on code completion, now boasting enhanced capabilities across both natural language and code. This dual proficiency makes it an invaluable asset for a wide array of AI workflows, including intelligent routing of requests, sophisticated question-answering systems, supporting sub-agents in multi-agent architectures, and facilitating private AI deployments within an organization's own infrastructure. The open-sourcing of Mellum2 underscores JetBrains' commitment to a "focal model" philosophy. This approach posits that while large, frontier models continue to push the boundaries of AI capabilities, practical and scalable AI products often benefit more from a coordinated system of specialized, faster models. These "focal models" are designed to efficiently handle high-frequency, latency-sensitive tasks that are common in modern AI applications, thereby complementing the more resource-intensive frontier models. JetBrains envisions Mellum2 playing a crucial role in the next generation of AI software tooling. Its applications are particularly relevant for developers building AI systems for software engineering, whether these are integrated into Integrated Development Environments (IDEs), utilized in Retrieval Augmented Generation (RAG) pipelines, or form part of complex agent workflows. By providing a high-performance, cost-efficient alternative, Mellum2 aims to empower developers to experiment, fine-tune, and deploy AI at scale within their own systems, fostering innovation and efficiency in AI-driven software development.