The open-source AI landscape just got a massive shot in the arm from an unexpected corner: JetBrains, the company behind IntelliJ IDEA and PyCharm, has open-sourced Mellum2 — a 12B parameter transformer model trained from scratch for real-world AI workflows.
What Is Mellum2?
Mellum2 isn't another chatbot or coding assistant. It's a workhorse model purpose-built for routing, Q&A, sub-agents, and private AI deployments inside software engineering systems. At just 12B parameters, it's lightweight enough to run on consumer hardware yet punches way above its weight on reasoning and instruction-following benchmarks.
Why This Matters
We've been so focused on frontier giants — GPT-5.6, Claude Opus 4.8, Gemini 3.2 — that we forgot the middle layer. Every production AI system needs routing models that decide which expert model to call, small sub-agents that handle focused tasks, and private inference where you can't send data to OpenAI or Anthropic.
Mellum2 fills that gap. JetBrains trained it from scratch on curated engineering data, optimized it for low-latency deployment, and released it under a permissive open-source license. You can download the weights today and run it locally or on your own infrastructure.
The Bigger Picture: Open-Source AI Tooling Is Booming
This isn't happening in isolation. The Open-Source AI Radar for June 2026 tracked 19 fast-rising GitHub repos — from model routers to agent orchestration frameworks to private deployment stacks. We're seeing a Cambrian explosion of tooling around open models, and Mellum2 is the latest, most polished example.
JetBrains knows developer tooling better than anyone. If they're betting on open-weight models for production AI workflows, every engineering team should pay attention.
Get the weights: JetBrains AI Blog
What open-source AI tool are you most excited about this month? Drop your thoughts below! 👇
Top comments (0)