JetBrains unveils Mellum, an open-source AI coding model
JetBrains, best known for its developer tools like IntelliJ IDEA and PyCharm, has taken a significant step in AI development. On Wednesday, the company released Mellum, its first openly available code-generating AI model, now accessible on Hugging Face. While Mellum has been in use within JetBrains’ development environments since last year, this release marks a shift toward broader community collaboration.
The model is trained on over 4 trillion tokens and has 4 billion parameters, making it robust for tasks like code completion. Tokens represent the raw input data such as lines of code while parameters reflect a model’s problem-solving ability. According to JetBrains, one million tokens equates to roughly 30,000 lines of code. Although Mellum isn’t plug-and-play, developers can fine-tune it for specific needs. JetBrains has already released fine-tuned versions for Python, though these are meant only for testing and experimentation, not production use.
Built for Developers, Researchers, and Educators
JetBrains designed Mellum for a range of coding-related applications. Whether integrated into intelligent developer environments (IDEs) or used as a foundation for research and education, it supports a wide spectrum of use cases. Additionally, the model is Apache 2.0 licensed, allowing for both academic and commercial reuse. It was trained over 20 days on a powerful cluster of 256 Nvidia H200 GPUs, using code from permissively licensed GitHub projects and content from English-language Wikipedia.
Though Mellum shows promise, it does come with limitations. The model must be fine-tuned before use, and its outputs may reflect biases found in open-source codebases. JetBrains also warns that code generated by Mellum may not be secure or vulnerability-free, especially since AI-generated code is already known to pose security risks. In fact, a 2023 Snyk survey revealed that over half of organisations face security issues with AI-assisted code.
A Starting Point for Future Collaboration
JetBrains isn’t aiming to create a one-size-fits-all solution. Instead, the company is focused on specialised, purpose-driven development. By releasing Mellum openly, it hopes to spark further experimentation and collaboration in the AI coding space. According to the company, even a single meaningful contribution from the community would be considered a success.
