Qwen3.8-Flash-Next
Qwen’s new open-weights model is a 125B-parameter multimodal MoE previewing Qwen4 architecture, with 6B active parameters.
Willison says he has been testing Qwen3.8-Flash-Next on a DGX Spark using Unsloth quantized builds. He has tried the 72.5GB UD-IQ1_S and 78.9GB UD-Q2_K_XL variants so far. His preferred result at the time of posting came from UD-Q2_K_XL with xhigh reasoning effort. Source: Simon Willison's note.
Willison says he has been testing Qwen3.8-Flash-Next on a DGX Spark using Unsloth quantized builds. He has tried the 72.5GB UD-IQ1_S and 78.9GB UD-Q2_K_XL variants so far. His preferred result at the time of posting came from UD-Q2_K_XL with xhigh reasoning effort. Source: Simon Willison's note.
score 8