JetBrains releases Mellum 2.1 for coding agents

JetBrains released Mellum 2.1, a 12-billion-parameter mixture-of-experts model oriented toward coding-agent workloads. The release extends the company's effort to integrate purpose-built language models into software-development tools.

A smaller specialized model can be useful for editor actions that need low latency and predictable cost. Model size alone says little about accuracy: agent reliability still depends on tool execution, repository context and evaluation on realistic programming tasks.