JetBrains local model needs a $2,700 Mac mini to run it

Apple just announced the Mac mini with M5 Pro and you’ll need one with 64GB of RAM to run the Junie Local model, a free 27B parameter coding agent that compares to Sonnet 4.5 or GPT-5.

JetBrains developed Junie as a coding agent that runs on your Mac and calls out to models the user chose. That always required some configuration and sending data off to some other AI product, so JetBrains has built Junie Local.

According to a blog post by JetBrains, Junie Local installs with a single command, takes about 20GB of space, and requires at least an M5 Mac with 64GB of RAM. Apple’s recently announced Mac mini costs $2,699 to meet that spec with an M5 Pro processor.

It runs using Qwen3.6-27B at 4-bit and is entirely local. There is no need for tokens or sending data to someone else’s server.

JetBrains targeted the M5 processor as the base spec due to the Neural Accelerator having 8-bit arithmetic instructions, which enables around 40% more prefill throughput. Qwen has a newer version, 3.8, but 3.6 was chosen since it’s faster without the “reasoning” required for 3.8.

The blog post suggests that JetBrains testing placed the 27B parameter coding agent on par with Sonnet 4.5 at a 10,000-token reasoning limit. GPT-5 at medium effort scored slightly higher.

Here’s what JetBrains had to say about the minimum spec requirements:

We know that an M5 Mac with 64 GB of RAM is a big ask. We are not going to pretend otherwise, and we know it puts Junie Local out of reach for many people reading the post.

That is simply what it costs to run a 27B model well today, and it is the number we are working hardest to bring down. The aim is a lower memory floor, wider hardware, and more of the stack optimized. If the lofty requirements are the reason you cannot try Junie Local, rest assured that we are working to bring them down.

Advertisement

JetBrains says that the Mac is a starting point, not the whole plan. It has started work for DGX Spark and RTX 5090, and they’ve looked at supporting 24GB cards.

Junie Local is free to install. Type /local into Junie and it installs. Switch back to cloud models using /model