# Best non-coding local model for edge device — retained record

> Retained reddit record from reddit:LocalLLaMA.

Canonical: https://highsignal.app/data/records/32e7b33f3d77a04a

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


source: [reddit](https://highsignal.app/data/reddit) (reddit:LocalLLaMA)  published 2026-10-25 07:37 ingested 2026-10-26 02:40

[original source ↗](https://www.reddit.com/r/LocalLLaMA/comments/1xew4bq/best_noncoding_local_model_for_edge_device/)

## retained text



```
Combining speed and size considered, what is the best general non-coding model for edge devices? I'm taking about mobile phone, SD 8g3 so the size should be around 4GB max. I tried Gemma 4 e2b, e4b, qat, q4, Qwen 3.5 4B, Bonsai 8B A1B.... None can answer accurate recipe viral last year, *Spaghetti all'assassina. All of them default to standard Bolognese and missing the charred*
```



document fetched 2026-10-26 00:38 · canonical https://reddit.com/r/LocalLLaMA/comments/1xew4bq/best_noncoding_local_model_for_edge_device
