JetBrains Unveils Mellum: An Open AI Model for Code Completion and Developer Tool Integration

JetBrains releases Mellum, an open AI coding model JetBrains, the creator of popular app development tools, has unveiled its first “open” AI model for coding. On Wednesday, the company made Mellum, a code-completion model previously used in its software development suites, freely available on the Hugging Face platform. Mellum, which boasts over 4 billion parameters and has been trained on more than 4 trillion tokens, is specifically tailored for code completion tasks, where it can suggest and complete code snippets based on the context provided by the developer. Parameters in AI models are akin to problem-solving skills, while tokens represent the raw data processed by the model. To put this into perspective, 1 million tokens are roughly equivalent to 30,000 lines of code. Mellum is designed to integrate seamlessly into professional developer tooling, such as intelligent code suggestions in integrated development environments (IDEs). It also holds promise for AI-powered coding assistants, research on code understanding and generation, and educational applications. Furthermore, the model is ideal for fine-tuning experiments, allowing developers to adapt it to their specific needs. Mellum is released under the Apache 2.0 license, making it freely accessible for both commercial and non-commercial use. During training, the model was fed a diverse range of datasets, including permissively licensed code from GitHub and content from English-language Wikipedia. The training process lasted approximately 20 days and was conducted on a cluster of 256 Nvidia H200 GPUs. However, getting Mellum operational requires some initial effort. The base model cannot be used out-of-the-box and needs to be fine-tuned before it can effectively assist in coding tasks. JetBrains has provided a few pre-fine-tuned models for Python, but these are intended only for assessing the model's potential capabilities and are not recommended for deployment in production environments. The advent of AI-generated code is transforming the software development landscape, but it also brings new security concerns. According to a late 2023 survey by the developer security platform Snyk, over 50% of organizations experience security issues with AI-produced code at least sometimes. JetBrains acknowledges these challenges and warns that Mellum might reflect biases found in public codebases, potentially generating code with similar styles to those in open-source repositories. Additionally, the code suggestions provided by Mellum are not guaranteed to be secure or free from vulnerabilities. Despite these caveats, JetBrains remains optimistic about the future of AI in coding. "This is just the beginning," the company noted in a blog post. "We’re not chasing generality—we’re building focus. If Mellum sparks even one meaningful experiment, contribution, or collaboration, we would consider it a success." By releasing Mellum, JetBrains hopes to foster innovation and collaboration within the developer community, empowering them to explore new possibilities in code generation and AI-assisted development.