JetBrains unveils Mellum, a specialized language model for code completion

JetBrains has launched Mellum, a purpose-built large language model engineered specifically for code completion, aiming to provide targeted assistance for developers. Unlike broad generalist models, Mellum is designed from the ground up to focus on a single core competency—delivering fast, lightweight, and high-quality code completion within developer workflows. JetBrains refers to this approach as a ´focal model,´ emphasizing depth in specialized domains rather than breadth across many tasks.

Mellum´s development marks the initial step in JetBrains´ strategic plan to craft a suite of focal models tailormade for distinct developer needs, such as diff prediction and potentially other coding-related tasks. The decision not to pursue a generalist Artificial Intelligence model underscores JetBrains´ belief that highly specialized models can achieve superior results in complex, high-value areas of software development. The company´s vision is to foster more precision, speed, and context-awareness in code-centric tasks by tightly aligning model capabilities with developer requirements.

Further promoting transparency and collaboration, JetBrains has open-sourced Mellum on the Hugging Face platform. This move invites researchers, engineers, and developers to experiment with the model, fine-tune it to their specific use cases, and contribute to its ongoing evolution. Mellum is now accessible both for integration within JetBrains IDEs and for open exploration, reflecting JetBrains´ commitment to community-driven advancement and the democratization of innovative Artificial Intelligence tools.