Qwen3.5-4B-3D-Coder vs Qwen3.8-27B

Official Qwen3.5-4B-3D-Coder Sakura Temple Garden Three.js scene screenshot from the ccore card — ccore/Qwen3.5-4B-3D-Coder card images/sakura-temple-garden.png. Source · License: Apache-2.0.

Verdict. Pick Qwen3.5-4B-3D-Coder if you want CCORE’s community 4B preview that writes Three.js scenes (GGUF + optional vision projector). Pick Qwen3.8-27B if you want Qwen’s official 27B native VLM (262K, SGLang/vLLM named). Both cards are Apache-2.0 and ungated. They are not the same weights and 3D-Coder is not an official Qwen release. We have no sourced bake-off.

Compare Qwen3.5-4B-3D-Coder and Qwen3.8-27B.

Dimension	Qwen3.5-4B-3D-Coder	Qwen3.8-27B	Source
Job	Community coding / Three.js HTML preview (optional image-guided GGUF)	Official native VLM (image + video + text in, text out)	Official cards / HF pipeline_tag
Official Qwen?	No (card: independent CCORE fine-tune)	Yes (Qwen/Qwen3.8-27B)	Official READMEs
License	Apache License 2.0	Apache License 2.0	Hugging Face LICENSE files
Gating	No	No	Hugging Face API
Developer	CCORE (community)	Qwen	Official cards
Base	Fine-tune of Qwen/Qwen3.5-4B	Official Qwen3.8 27B lead card	HF cardData / official card
Parameters	4B (card); safetensors 4,539,265,536 F16	27B (card); safetensors 27,781,427,952	Official READMEs / HF API
Context	262,144 (config); llama.cpp example 65,536	262,144 native; card names 1,000,000 extension	config.json / official READMEs
Hugging Face likes (lead)	478 (2026-11-02)	18,393 (2026-10-26)	Hugging Face API
Quality vs ChatGPT / Claude Code	Not independently tested	Not independently tested	n/a

When to choose which

Interactive Three.js / voxel HTML scenes: Qwen3.5-4B-3D-Coder — Community specialty card. Not a quality rank.
Official Qwen 27B native VLM: Qwen3.8-27B — Official weights. 3D-Coder is not a Qwen release.
Small local GGUF coding preview: Qwen3.5-4B-3D-Coder — Q4_K_M / Q8_0 siblings on the same repo. 27B is a different size class.
Same Apache-2.0 weight grant: Tie — Both LICENSE files are Apache License 2.0. Not legal advice.

Read the entity pages: Qwen3.5-4B-3D-Coder and Qwen3.8-27B. 3D-Coder is community, not official Qwen. Both hang off ChatGPT alternatives and closed LLM APIs. 3D-Coder also hangs off Claude Code alternatives and Cursor alternatives.

Why this comparison exists

The 4B card is named Qwen3.5-4B-3D-Coder. The official 27B card is named Qwen3.8-27B. People search those strings against each other. This page keeps them distinct with sourced rows only. It is identification, not a bake-off.

Same-lab image-to-3D (different modality) stays on Idol / Meshy alternatives. Another Qwen-family code-driven card already here is SenseNova 6.8 Code2Video (video, not Three.js). ~4B on-device coding that is not Qwen-named: Spark-X2.6-4B.

Sourced snapshot (2026-11-02 / 2026-10-26)

Field	Qwen3.5-4B-3D-Coder	Qwen3.8-27B
Host	huggingface.co/ccore/Qwen3.5-4B-3D-Coder (not gated)	huggingface.co/Qwen/Qwen3.8-27B (not gated)
HF likes / downloads	478 / 44,825 (2026-11-02)	18,393 / 7,762,014 (2026-10-26)
License	Apache License 2.0	Apache License 2.0
Official Qwen	No	Yes
Size	4B (safetensors 4.54B F16)	27B (safetensors 27.78B)
Job	Coding / Three.js HTML	Native VLM (image+video+text)
Context	262,144 (config)	262,144 native; card names 1M
Local path named	FP16 + Q4_K_M + Q8_0 + optional mmproj	SGLang / vLLM / TokenSpeed
Quality vs ChatGPT	Not independently tested	Not independently tested

License

Both LICENSE files are Apache License 2.0. Card tags apache-2.0. Hosted Qwen Cloud terms are separate from the 27B weight grant. Not legal advice.

How to choose

Need a community 4B Three.js / coding preview with GGUF: Qwen3.5-4B-3D-Coder.
Need official Qwen 27B VLM: Qwen3.8-27B.
Need another official-ish Qwen3.8-27B fine-tune (code-driven video): SenseNova 6.8 Code2Video.
Need Apache-2.0 ~4B that is not Qwen-named: Spark-X2.6-4B.
Need image-to-3D splat, not an LLM: Idol on Meshy alternatives.
Engine page: vLLM.

See also Dev & Coding and Models. Catalog: home. Editorial: About. Corrections: contact.

FAQ

Which one should I download first?

Need a small community Three.js / coding preview with GGUF — 3D-Coder. Need official Qwen 27B vision-language weights — Qwen3.8-27B. Do not treat the 4B card as a distilled official Qwen3.8.

Is 3D-Coder official Qwen?

No. The CCORE card says it is an independent fine-tune of Qwen3.5-4B and “is not an official Qwen release.” Qwen3.8-27B is the official 27B card already on this site.

Why this pairing instead of vs Spark-X2.6-4B or vs Idol?

Searchers already confuse Qwen-named cards. Spark-X2.6-4B is a different ~4B coding/agent line (not Qwen). Idol is an image-to-3D Gaussian splat, not an LLM. Those stay on their own pages.

Does 3D-Coder replace Claude Code?

No sourced bake-off. See Claude Code alternatives. 3D-Coder is weights, not a coding agent.

Sources & verification

Last verified 2026-11-02 · Last updated 2026-11-02· Edited by forkalt editors · Editorial policy

License notes are not legal advice. Read the original LICENSE. Trademark & disclaimer.