# Nvidia Nemotron 3.5 Super 120B vs. Qwen 3.8 27B - comparison of storytelling and visual animations in html — retained record

> Retained reddit record from reddit:LocalLLaMA.

Canonical: https://highsignal.app/data/records/5da1663634f5e9e8

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:LocalLLaMA)  published 2026-10-20 08:22 ingested 2026-10-21 02:41

[original source ↗](https://www.reddit.com/r/LocalLLaMA/comments/1xamhwt/nvidia_nemotron_35_super_120b_vs_qwen_38_27b/)

## retained text



```
I was testing the visualization capabilities of different models and came up on quite a dramatic difference between Nemotron and Qwen, wanted to share, just as a difference to the "draw a pelican riding a bicycle" test.

Prompt: *"Write a single html animation with canvas and no libraries explaining visually how noise cancelling headphones work. Plan out the explanation steps and storyboard them with the text, labels, and visuals, then do the coding. Visually, use frosted-glass UI: translucent panels that genuinely blur the vivid gradient blobs and orbs behind them, 1px light borders, diagonal specular sheen, fine frosted grain and crisp inter type."*

Results (model produced html page as result, these are videos of the resulting animation)

**NVidia Nemotron 3.5 Super VL 120B A12B Q4_K_M** (thought for 7 minutes, 31 tps)

https://imgur.com/B3l0qV1

**Qwen 3.8 27B Q8_0** (thought for 20 minutes, 15 tps)

https://imgur.com/NsKj4XR

Tested on AMD Strix Halo 128GB, LM Studio on Win11.

It's pretty clear which model gave a better result! It really hits way above its weight class.
```



document fetched 2026-10-21 00:41 · canonical https://reddit.com/r/LocalLLaMA/comments/1xamhwt/nvidia_nemotron_35_super_120b_vs_qwen_38_27b
