Qwen 3 Coder Next: LM Studio and Zed setup and tests

Nick tests Qwen 3 Coder Next as a coding assistant, using the 80 billion parameter model with a download size of roughly 50 GB. His Linux desktop has an AMD Ryzen 7 CPU, 128 GB RAM and a GeForce RTX 4060 Ti with 16 GB VRAM. He reports responsive generation on this setup; the demonstration does not establish minimum hardware requirements.

LM Studio runs the model on the desktop and exposes its built-in API over the home network. Nick connects Zed on his MacBook by entering the desktop's network address and server port in the editor's LLM provider settings. NVTOP monitors GPU utilization and memory use, which approaches the card's VRAM capacity after loading.

The local LLM completes a simple Python function, but gets stuck on a request for one HTML file visualizing six sorting algorithms. Nick says Gemini 3.1 and a cloud-hosted 400 billion parameter Qwen 3.5 completed that task in earlier tests. Those comparisons involve different models and deployment conditions.

A smaller request for a single sorting visualizer produces HTML with embedded CSS and JavaScript. Nick fixes four misplaced closing tags before testing it. He then reports that sorting, reset and the other controls work. His verdict is that the model can handle small and medium coding tasks, while more complex work benefits from smaller subtasks. He does not consider it a full replacement for paid models.