SGLang reports Qwen3.8 Flash Next runs up to 6.6× faster on Vera Rubin than GB300

SGLang says Qwen3.8 Flash Next on NVIDIA Vera Rubin, running with NVIDIA Dynamo and SGLang, delivers up to 6.6× throughput per chip versus GB300, and up to 76% lower modeled cost per total token at matched per-user speed.

SGLang credits NVIDIA for partnership and early access, and SemiAnalysis for the results. SemiAnalysis's separate analysis puts Rubin at 4.23× performance per dollar over GB300 NVL72 on SGLang.

SGLang also shares a link to an InferenceX comparison of Qwen3.8-Flash-Next on GB300 and VR200. The linked page compares per-GPU throughput under agentic-trace workloads at FP4 precision.