JetBrains Releases Mellum 2.1: 12B MoE Model Optimised for Coding Agents

Mellum 2.1 represents a focused approach to building specialist models for local deployment. As a 12B Mixture-of-Experts model optimised for coding, it strikes a practical balance between capability and resource consumption. MoE architectures enable selective activation of model parameters during inference, allowing effective capability of larger models while maintaining the latency and memory profile of smaller dense models—ideal for edge and on-device scenarios.

The emphasis on coding agents reflects the maturation of local LLM applications. Rather than pursuing general-purpose capabilities, specialised models like Mellum 2.1 can deliver superior performance on well-defined tasks while remaining deployable on modest hardware. For developers building coding assistants, IDE integrations, or autonomous development agents, open-source MoE models eliminate dependency on proprietary APIs and enable fine-tuning and customisation for domain-specific requirements.

Read the full article on Google News.

Source: Google News · Relevance: 8/10