12B MoE

JetBrains

Mellum 2 — open 12B MoE coding models, in Thinking and Instruct flavours.

Mellum 2 is JetBrains’ family of open-source language models built for coding and language tasks. Optimised for low-latency, high-throughput inference, Apache 2.0 licensed, and deployable locally or in the cloud — for coding assistants, RAG apps, code analysis and developer tools.

The kit, sized small

Mellum 2 Thinking

Reasoning-heavy configuration for harder problems.

  • coding
  • reasoning
12B parameters32B cap
Mellum 2 Instruct

Blazingly fast instruct configuration for high-throughput use.

  • coding
  • low-latency
12B parameters32B cap

If you want to build…

An AI coding assistant
Mellum 2 InstructLow-latency
Reasoning-heavy code tasks
Mellum 2 ThinkingDeeper reasoning
RAG or code-analysis tools
Mellum 2Apache 2.0