This week a Reddit developer pulled off something we think deserves a closer look: he ran Alibaba's Qwen3 27B (a locally deployable open-source large model) on a single RTX 5090, prompting the AI to write code that drives the CAD tools cadquery and Blender, design a 3D-printable self-watering planter and a MagSafe phone stand, then generate product ad imagery with Qwen Image 2.1. The entire pipeline runs in his home on hardware totaling under ¥20,000.

The most striking detail: the "skills" code that orchestrates this stack — telling the model how to use CAD, how to use ComfyUI — was written by another AI assistant called Pi, not by a human.

What this is

This is a concrete demonstration of open-source large models moving "from chatting to actually working." Until now, the "is it usable" question basically meant "can it write an article." This time Qwen3 27B completed five real engineering steps — modeling, writing code, calling tools, rendering, compositing — and the output goes straight into a printer. This isn't a one-off breakthrough: over the past six months, open-source models have been steadily accumulating capability in long-chain tool use (letting a model autonomously invoke external software step by step to complete a task), but the news mostly covers benchmark scores, with few people running an entire pipeline end-to-end and showing the finished product.

Industry view

The optimistic read: open-source models have crossed the threshold of "running real work," and local AI is no longer just a chat toy. For small-batch manufacturers that don't want to ship drawings to the cloud, this is good news.

But the editorial team also notes several caveats. First, this was a hobbyist setup on Reddit with an AI-generated toolchain — it can't be copy-pasted into a production line; cadquery's geometric output often needs manual cleanup before it's actually printable. Second, there's still a gap between "can run through one planter" and "can reliably produce manufacturable parts" — the poster himself says the next thing he wants to try is "recreating 3D models from photos," which is a different mountain. Third, a 5090 plus a 3080 is still a configuration most households won't pay for.

Impact on regular people

For enterprise IT: Small and mid-size manufacturers that don't want to upload design files to the cloud but do need custom parts should at least evaluate piloting local open-source models plus consumer GPUs as an "internal AI design assistant" — minimum viability is now in reach.

For individual professionals: Product designers and indie developers can build an AI workflow that produces usable output for under ¥20,000, which may substitute for some outsourced rendering or template purchases.

For the consumer market: If 3D printing plus AI design keeps maturing, "custom small objects" — currently a niche path — will enter more homes; but copyright and usability-compliance issues around generated content will also surface.