CCORE releases final 4B model that generates 3D assets as Three.js scenes with GLB export, weights on Hugging Face
ConfidentDinner6648 · reddit · 2026-11-05
The final version of CCORE GLB-3D-Coder is out: a full fine-tune of Qwen3.5-4B that turns text descriptions into complete Three.js HTML scenes with an Export GLB button, ready for Godot, Blender, or Unity. It covers characters, buildings, weapons, vehicles, furniture and more, and can bake animations into the GLB. Available as Q4KM GGUF (2.7 GB), Q80 GGUF, and full Safetensors weights on Hugging Face; the training dataset releases November 6. The author shows side-by-side comparisons against DeepSeek V4.1 Flash, highlighting the detail achieved at only 4B parameters.
More from Models
- Developer claims visual memory works surprisingly well on GPT-4o, encoding data into AI-generated images — RileyRalmuto · 2026-11-05
- Gemini Omni 1.2 Flash Leak Spotted in API Quota Dashboard, Release Imminent — koltregaskes · 2026-11-05
- Gemini 4 Carbon tops Artificial Analysis; ThursdAI covers S-1 charts, Genie 4 audio — altryne · 2026-11-05
- 'Copyright-Claude' meme: a demon summoned by optimizing under conflicting constraints — layer07_yuxi · 2026-11-05
- Latent Space: OpenAI pipeline patches 948 Linux bugs; Kimi K3.1 Pro, SWE-2 Flash, Astra sandbagging — Latent Space · 2026-11-05
- Users report ChatGPT Codex, Gemini and DeepSeek all degrading sharply since early October — Alex-2503 · 2026-11-05