twenty sixth August 2026 – Hyperlink Weblog
Qwen3.8-Flash-Subsequent (by way of) One other open weights mannequin from Qwen. This one is “a multimodal MoE mannequin that additionally serves as an early preview of the structure utilized in Qwen4”.
It is fairly massive: 125B tokens, however solely 6B energetic which suggests it will get a major efficiency enhance.
I have been attempting it out on a DGX Spark utilizing these Unsloth quantized fashions. I am nonetheless exploring the mannequin – thus far I’ve tried the 72.5GB UD-IQ1_S one (producing these pelicans) and the 78.9GB UD-Q2_K_XL (producing these).
My favourite thus far was this xhigh reasoning effort one from UD-Q2_K_XL:
