Qwen 2.5 Coder 32B Runs on a Single 24GB GPU

Running a serious coding model at home usually means two graphics cards and a lot of patience. Qwen 2.5 Coder 32B does it on one. It is the third pick in a series on open-weight models you can run yourself, and it has become the local coding standard.

It ships under Apache 2.0, so commercial use is fine, and it takes 128k tokens of context natively.

What Happened

Qwen 2.5 Coder 32B is a software engineering specialist released under the Apache 2.0 license. The recommendation also notes it stays easygoing when a task looks like security testing, which matters if you put it to that kind of work.

Here is what it wants from your machine:

• VRAM: one RTX 3090 or 4090 (24GB) is enough for Q4 or Q5 quants. Q8 needs two 24GB cards.
• System RAM: 32GB to 64GB.
• Storage: about 35GB for the GGUF Q5_K_M file.
• Context window: 128k tokens, native.

Why It Matters

That is a bill of materials you may already own. A single 24GB card, 32GB to 64GB of system memory and 35GB of free disk is the whole requirement, which puts a 32B coding model on prosumer hardware instead of a rented server.

If you are shopping, the one spec that decides everything is the 24GB of VRAM. If you already have a 3090 or 4090, the only thing left to clear is the 35GB of storage.