Qwen 3.8 Flash
An efficient visual language model released by Alibaba with approximately 180 billion parameters. It natively supports visual multimodal capabilities and is suitable for local deployment, demonstrating strong performance in inference speed and resource efficiency.
Timeline (last 90 days)
v3.0.68 switches GMI Cloud's default model to Qwen 3.8 Flash.
The user claims that this DDR3 multi-socket server configuration can run Qwen 3.8 Flash at over 15 tok/s
Under a dual-Spark TP2 configuration, Qwen 3.8 Flash achieves an inference speed of 55–70 tokens/s and supports 4 fully concurrent agents running simultaneously.
The model showed no speed advantage in TP4 mode; measured performance fell between TP1 and TP2, with TP2 considered the gold-standard configuration.
2×RTX 3090 can run Qwen 3.8 Flash, but requires substantial system memory support.
The author predicts that if the Qwen team releases a Flash version optimized for TP4, inference speed on DGX Spark could be boosted to the 100 tokens/s level.
The quantized Qwen 3.8 Flash performs stably in scenarios such as building HTML games, web search, and browser automation tasks.
Qwen 3.8 Flash has a parameter count of 180B, which is half that of GLM 5.3 Flash (320B).
The daily free token quota for the Qwen 3.8 Flash model on the platform is approximately 10 billion tokens.
All Facts (6)
Qwen 3.8 Flash has a parameter count of 180B, which is half that of GLM 5.3 Flash (320B).
50%UnverifiedUnder a dual-Spark TP2 configuration, Qwen 3.8 Flash achieves an inference speed of 55–70 tokens/s and supports 4 fully concurrent agents running simultaneously.
50%UnverifiedThe model showed no speed advantage in TP4 mode; measured performance fell between TP1 and TP2, with TP2 considered the gold-standard configuration.
50%UnverifiedThe quantized Qwen 3.8 Flash performs stably in scenarios such as building HTML games, web search, and browser automation tasks.
50%Unverified2×RTX 3090 can run Qwen 3.8 Flash, but requires substantial system memory support.
50%UnverifiedThe author predicts that if the Qwen team releases a Flash version optimized for TP4, inference speed on DGX Spark could be boosted to the 100 tokens/s level.
50%