Qwen3.8-Flash-Next Hits 685 Output Tokens/s per User on NVIDIA GB300

Key Info

LightSeek's TokenSpeed benchmark reports Qwen3.8-Flash-Next reaching 685 output tokens/s per user, running on 4x NVIDIA GB300 GPUs with NVFP4.

Highlights

• 685 output tokens/s per user on Qwen3.8-Flash-Next
• Deployed on 4x NVIDIA GB300 with NVFP4 precision
• Kernel and serving optimizations are documented in a linked write-up
• Qwen highlights community efforts to unlock inference potential