JetBrains has launched Mellum, a proprietary AI model crafted specifically for software development, now available through JetBrains’ AI Assistant. Mellum’s integration has already yielded significant advancements in the speed and accuracy of code completions compared to previous models.

Unlike larger language models, Mellum is purpose-built with a lightweight architecture to deliver near-instant coding suggestions. Currently, it supports five key programming languages—Java, Kotlin, Python, Go, and PHP—with early access to additional languages in progress.

Since Mellum’s introduction, JetBrains has observed remarkable performance improvements. Code completion latency has been reduced to one-third of its previous duration, and the acceptance rate for Mellum’s suggestions has reached an industry-competitive 40%. Furthermore, there has been a 3-4 times decrease in canceled suggestions, and the number of completions offered has more than doubled.

A Strategic Leap for JetBrains AI Assistant

Vladislav Tankov, JetBrains’ Director of AI, explained, “The JetBrains AI Assistant selects the most suitable model from providers like OpenAI and Google, but to maximize our code completion capabilities, we needed our own model. The Mellum model represents a major step forward for us.”

Tankov highlighted that the model’s strength comes from its deep integration with JetBrains IDEs, combining client-side integration with server-side logic to deliver swift, accurate, and context-aware coding suggestions. This synergy enhances developers’ coding experience with outstanding results.

JetBrains has also prioritized data privacy in Mellum’s design. The model is trained solely on publicly available code with permissive licenses, and the company confirms that it neither stores nor uses any customer input during operation.