Mellum2 Goes Open Source: A Fast Model for AI Workflows
4 hours ago
- #Software Engineering
- #AI
- #Open Source
- JetBrains has open-sourced Mellum2, a 12B parameter Mixture-of-Experts model designed for high-performance, low-latency AI workflows in software engineering.
- Mellum2 specializes in natural language and code, enabling use cases like workload routing, RAG pipelines, sub-agent tasks, and private local deployments.
- With only 2.5B active parameters per token, the model reduces compute costs and inference time by over half compared to similar-sized models.
- JetBrains advocates for a 'focal model' philosophy, using specialized, efficient models within coordinated AI systems to address scalability and latency challenges.