Skip to content
4All API Docs
Main Site News Console

Qwen3.8-Flash-Next

· Simon Willison Translated
技巧LLM

Qwen3.8-Flash-Next

Another open-weight model from Qwen. It is a “multimodal MoE model and an early preview of the architecture used by Qwen4.”

It is fairly large, containing 125B tokens, but activating only 6B at a time, which enables significant performance gains.

I’ve been testing these models quantized by Unsloth on a DGX Spark. I’m still exploring the model—so far, I’ve tried the 72.5GB UD-IQ1_S version (which generated these pelicans) and the 78.9GB UD-Q2_K_XL version (which generated these images).

So far, my favorite is this image generated by UD-Q2_K_XL with xhigh reasoning effort:

Flat vector illustration: a white pelican with an orange beak and orange legs riding a red bicycle along a sandy path. A blue fish sits in a wicker basket on the handlebars. The background features rolling green hills, a small tree and bushes, white clouds, and a bright yellow sun in a blue sky.