Qwen3.8-Flash-Next is a multimodal MoE model with 125B total and 6B active parameters, serving as an early preview of the architecture planned for Qwen4. The author tested Unsloth quantized versions on a DGX Spark, including the 72.5GB UD-IQ1_S and 78.9GB UD-Q2_K_XL builds, generating image outputs from each. The author's preferred result so far came from an xhigh reasoning effort run using UD-Q2_K_XL.
No score is assigned. Sources and their independence are shown in the citation chain below.