Overview / Description
Mellum by JetBrains is an AI developer tool, a family of fast language models purpose-built for code completion in JetBrains IDEs, with an open-source 4-billion-parameter base model released on Hugging Face. JetBrains describes Mellum as a "focal" model built from scratch for one job, low-latency cloud-based code completion, rather than a general-purpose chatbot, prioritizing high-performance inference for real coding workflows. The Mellum-4b-base model supports completion across Java, Kotlin, Python, Go, PHP, C, C++, C#, JavaScript, TypeScript, CSS, HTML, and Rust, giving broad multilingual coverage for mixed codebases. Because the base model is fully open-sourced and freely downloadable, it is aimed at AI/ML researchers, engineers, and educators who want to build tooling, run experiments, or fine-tune a code model, not only end users clicking autocomplete in an IDE. As an open code-completion LLM, Mellum trades broad chat ability for speed and focus on the developer autocomplete task. The base model is available at no cost on Hugging Face; JetBrains does not publish separate pricing for the open weights.
Used For
Fast multilingual code completion and code-model research for developers and AI/ML engineers
Pricing
Pros & Cons
Pros
- Purpose-built 4B model focused on code completion, not general chat
- Open-sourced Mellum-4b-base freely available on Hugging Face
- Multilingual: Java, Kotlin, Python, Go, PHP, C/C++, C#, JS, TS, CSS, HTML, Rust
- Optimized for low-latency, high-performance inference in coding workflows
- Suitable for research, fine-tuning, and building custom developer tooling
Cons
- Narrow scope: a 'focal' completion model, not a full chat or agent LLM
- Aimed at AI/ML researchers and engineers, not plug-and-play for casual users
- Requires infrastructure to self-host and run the open weights
- License details are not spelled out on the product page
Questions & Answers
Alternatives
StarCoder, Code Llama, CodeGemma, Codestral