AIAny
AI Model2026
Icon for item

Qwen3.5-4B-3D-Coder

Turns natural-language descriptions into runnable Three.js scene code, exploring how a small language model can generate structured interactive 3D environments as well as general-purpose code.

Introduction

A useful test of a small coding model is not just whether it writes valid JavaScript, but whether that code produces a coherent space. This preview focuses on that harder step: turning detailed spatial prompts into interactive Three.js scenes, with a 4B-parameter model rather than a large general-purpose system.

Key Capabilities
  • Scene generation, not image generation: It produces HTML and JavaScript for rendered 3D environments, so the output can be inspected and modified as code.
  • Spatially detailed prompts: Examples ask for connected structures, paths, elevations, and functional relationships—not just a collection of visual objects. This makes it useful for quickly prototyping scene layouts.
  • General coding alongside 3D: The creator reports promising early results beyond scene creation, though the project’s main emphasis is Three.js.
  • Still an early preview: October checkpoint updates report improved one-shot results, but some generations still need retries or corrections; duplicate variable names are a known issue.
Who It Suits

Great fit for developers and 3D hobbyists who want a local, small-model starting point for scene prototypes and are comfortable checking and revising generated code. Look elsewhere if you need consistently production-ready scenes, validated coding benchmarks, or a stable official Qwen release: this is an independent fine-tune still under development.

Information

  • Websitehuggingface.co
  • OrganizationsCCORE, CCORE Desenvolvimento
  • AuthorsCaio Giovannetti
  • Published date2026/10/25

More Items

Hugging Face
AI Model2026

Translates incoming Chinese or English speech incrementally, aiming to keep lag low without losing meaning. This FP8 checkpoint returns translated text; speech synthesis and voice cloning require a separately configured TTS service.

Hugging Face
AI Model2026

Runs local agents, tool calls, and structured-output tasks with a 256K-token context window. At 4-bit precision it uses about 1 GB of memory, and it can be fine-tuned for specialized domains on a single GPU.

GitHub
AI Coding2026

Provides reusable design rules for AI coding tools, covering page structure, copy, and visual styling. Skills are distributed as standalone Markdown files that work across multiple coding assistants.