# Introducing Mellum2: A 12B Mixture-of-Experts Model by JetBrains

Published: 2026-06-01 · Updated: 2026-06-01 · Category: Model Release

Source page: https://www.sipply.dev/story/introducing-mellum2-a-12b-mixture-of-experts-model-by-4b24f824 · 中文: https://www.sipply.dev/zh/story/introducing-mellum2-a-12b-mixture-of-experts-model-by-4b24f824

## Summary

JetBrains has unveiled Mellum2, a new Mixture-of-Experts model featuring 12 billion parameters. This innovative model activates only a fraction of its parameters—2.5 billion per token—allowing it to deliver over twice the inference speed of comparable models. This efficiency makes Mellum2 particularly well-suited for tasks in natural language processing and code generation, catering to the needs of software engineers and developers alike.

## Key takeaways

- Mellum2 is a 12B parameter model that uses a Mixture-of-Experts architecture.
- The model activates only 2.5B parameters per token, significantly enhancing inference speed.
- Mellum2 is designed to improve efficiency in natural language and code processing tasks.

## Why it matters

The introduction of Mellum2 represents a significant advancement in AI model architecture, particularly for applications in software engineering. Its efficiency could lead to faster development cycles and more responsive AI applications.

- **Product managers:** For product managers, Mellum2's speed and efficiency can translate into quicker iterations and improved user experiences in software products that leverage AI capabilities.
- **Developers:** Developers can benefit from Mellum2's enhanced performance, allowing them to integrate AI functionalities into their applications more effectively and with reduced latency.
- **Students and job seekers:** Students and job-seekers in the AI and software development fields should take note of Mellum2 as it exemplifies cutting-edge technology that they may encounter in future projects and job opportunities.

## Sources

- [Introducing Mellum2: A 12B Mixture-of-Experts Model by JetBrains](https://huggingface.co/blog/JetBrains/mellum2-launch) (Hugging Face Blog, 2026-06-01)

## 中文

### 介绍Mellum2：JetBrains推出的12B专家混合模型

JetBrains推出了Mellum2，这是一种新的专家混合模型，具有120亿个参数。该创新模型每个token仅激活部分参数——每个token激活25亿个参数——使其推理速度超过同类模型的两倍。这种高效性使Mellum2特别适合自然语言处理和代码生成任务，满足软件工程师和开发人员的需求。

**要点：**

- Mellum2是一个具有120亿参数的模型，采用专家混合架构。
- 该模型每个token仅激活25亿个参数，显著提升推理速度。
- Mellum2旨在提高自然语言和代码处理任务的效率。

**为什么重要：** Mellum2的推出代表了AI模型架构的重大进步，特别是在软件工程应用方面。其高效性可能导致更快的开发周期和更具响应性的AI应用。

---
Sipply: Sip AI News Simply. https://www.sipply.dev