JetBrains Junie Local Runs Qwen3.6 Directly on M5 Macs

JetBrains has launched Junie Local, a new version of its AI coding agent that runs entirely on a Mac instead of sending coding work to the cloud. The company says the first release is free, uses a tuned Qwen3.6-27B model, and keeps prompts, source code, and diffs on the machine.

Junie Local is designed to remove the manual setup normally required for local AI coding. JetBrains says users can open Junie and run /local, after which the model downloads, a local server starts, and the agent switches to the on-device model. JetBrains' announcement says no separate runtime or configuration file is required.

Junie Local Brings a Tuned Qwen Model to the Mac

The bundled model is Qwen3.6-27B in 4-bit form, requiring about 20GB of downloads. JetBrains says Junie Local currently needs an M5 Mac with 64GB of RAM, putting the first release firmly in the high-end Mac category.

JetBrains says it chose Qwen3.6 rather than the newer Qwen3.8 because its tests found that Qwen3.8's reasoning mode made coding tasks substantially slower. The company also optimized prefill and added techniques such as KV-cache reuse and speculative decoding to improve local performance.

Saganote recently covered Qwen3.8-Max, giving readers context on the newer Qwen generation that JetBrains says it did not select for Junie Local.

The Coding Agent Works Without Cloud Credits

Junie Local is not simply a local chatbot. JetBrains says the same Junie agent experience remains available, including plan mode, live prompting, guidelines, skills, and commands. The difference is that the model runs on the user's hardware instead of consuming cloud tokens or credits.

That makes the setup useful for repetitive engineering work such as multifile refactors, dependency upgrades, framework migrations, test coverage, and understanding an inherited codebase. JetBrains also says the agent can operate without a network connection once the model is stored locally.

Saganote's GitHub Copilot, Cursor, and Tabnine comparison covers the wider AI coding assistant market, while Meta's Muse Code shows how coding agents are increasingly moving beyond autocomplete toward larger software tasks.

Junie Local Starts With High Hardware Requirements

The biggest limitation is the hardware requirement. JetBrains acknowledges that an M5 Mac with 64GB of RAM excludes many Mac users, but says it is working toward a lower memory requirement and broader hardware support.

The company says Mac is only the starting point. It already has prototypes running on DGX Spark and RTX 5090 hardware and is evaluating systems with 24GB graphics cards.

For now, Junie Local is available at no cost, with no registration, subscription, credits, or payment card required. The launch gives developers a way to use an agentic coding workflow locally while keeping their project data on their own hardware.