Built this yesterday with Qwen3.8-Flash-Next (NVFP4, 262K context) on a single NVIDIA DGX Spark

A developer built an application in 8 hours using Qwen3.8-Flash-Next (NVFP4, 262K context) on a single NVIDIA DGX Spark. The project involved generating approximately 10k lines of code and consuming around 800k tokens, with VSCode Copilot in autopilot mode and SGLang. The developer noted that while it's not "GPT-6 Astra level," achieving around 35 tok/s with a 100% local ~180B MoE on a single DGX Spark is a significant accomplishment.

This report highlights the performance of Qwen3.8-Flash-Next, showing it can achieve ~35 tok/s with a ~180B MoE model on a single DGX Spark, unlike other models that typically require more extensive hardware.