
JetBrains Releases Mellum, an ‘Open’ AI Coding Model
Today, JetBrains announced the release of Mellum, an open-source AI coding model designed for code completion and research purposes. According to the company, Mellum is trained on more than 4 trillion tokens and has 4 billion parameters, making it a powerful tool for developers and researchers alike.
Mellum’s capabilities include generating code snippets based on context, which can greatly enhance productivity in software development suites. The model can also be used in educational settings to help teach coding skills or as a fine-tuning experiment for other AI models.
It is worth noting that JetBrains has trained Mellum on publicly available data sets, including permissively licensed GitHub code and Wikipedia articles. This raises concerns about the potential biases present in public codebases and the security implications of using AI-generated code. As such, JetBrains warns users that the model may “reflect biases present in public codebases” and that generated code suggestions won’t necessarily be secure or free of vulnerabilities.
Despite these limitations, JetBrains sees Mellum as a major step forward in the development of AI-powered coding assistants. In a blog post announcing the release, the company stated that they are not chasing generality but rather building focus. If Mellum sparks even one meaningful experiment, contribution, or collaboration, they would consider it a win.
Mellum is currently available on Hugging Face’s AI dev platform and can be fine-tuned for specific languages such as Python. However, JetBrains cautions that the base model cannot be used out of the box and requires initial fine-tuning before deployment.
Source: https://techcrunch.com/2025/04/30/jetbrains-releases-mellum-an-open-ai-coding-model/