JetBrains Launches Mellum: An Open AI Model for Code Generation

JetBrains, the renowned developer behind popular coding tools like IntelliJ IDEA, has unveiled Mellum, its first openly accessible AI model for code generation. Now available on Hugging Face, Mellum is designed to assist developers with intelligent code suggestions, research, and educational applications.

Trained on 4 trillion tokens—equivalent to roughly 120 billion lines of code—Mellum boasts 4 billion parameters, enhancing its ability to understand and generate code snippets. “Designed for integration into professional developer tooling, AI-powered coding assistants, and research on code understanding and generation, Mellum is also well-suited for educational applications and fine-tuning experiments,” JetBrains explained in a technical report.

The model was trained on a mix of permissively licensed GitHub code and English Wikipedia articles, taking approximately 20 days on a cluster of 256 Nvidia H200 GPUs. However, Mellum isn’t plug-and-play—developers must fine-tune it for specific use cases. JetBrains has provided pre-tuned versions for Python but warns they’re for experimental purposes only, not production deployment.

While AI-generated code accelerates development, it also raises security concerns. A 2023 survey by Snyk found that over 50% of organizations face security issues with AI-produced code. JetBrains acknowledges Mellum may inherit biases from public codebases and doesn’t guarantee vulnerability-free suggestions.

“This is just the beginning. We’re not chasing generality—we’re building focus,” JetBrains stated. The company hopes Mellum will inspire collaboration and innovation in the AI coding space.