Wesche Tests Qwen3.8 Flash on DGX Sparks

Wesche, who builds and benchmarks frontier LLMs, posted a video showing tests of Qwen3.8 Flash Next in NVFP4 format. The post describes running the model across multiple users with independent prompts and KV caches while checking context handling at scale. The author notes the exercise measures hardware limits on the Spark systems rather than offering a production setup and tags the post to NVIDIA AI.