Quen 3.8 Flash Next
A large-scale open-source mixture-of-experts (MoE) language model from Alibaba, with a total of 125 billion parameters and 6 billion parameters activated per token, designed for local deployment on consumer-grade hardware at a relatively high level of intelligence.
Timeline (last 90 days)
Doug Ware tested Alibaba's open-source model Quen 3.8 Flash Next (a 125-billion-parameter MoE model that activates 6 billion parameters per token) within two weeks on a Strix Halo mini PC with 128GB of memory and an NVIDIA CUDA workstation.
Quen 3.8 Flash Next surpasses GPT-6 Luna and Cloud Haiku 5.5 on benchmark scores.
Loading the selected version of Quen 3.8 Flash Next requires 111GB of memory, whereas the dense Quen 3.8 27b model requires only 23GB.
All Facts (3)
Quen 3.8 Flash Next surpasses GPT-6 Luna and Cloud Haiku 5.5 on benchmark scores.
50%UnverifiedLoading the selected version of Quen 3.8 Flash Next requires 111GB of memory, whereas the dense Quen 3.8 27b model requires only 23GB.
50%UnverifiedDoug Ware tested Alibaba's open-source model Quen 3.8 Flash Next (a 125-billion-parameter MoE model that activates 6 billion parameters per token) within two weeks on a Strix Halo mini PC with 128GB of memory and an NVIDIA CUDA workstation.
50%