Skip to content
Trending storyDeveloping

Qwen3.8-Flash-Next reaches 685 output tokens/s per user on GB300 via TokenSpeed

1 article1 sourcesince Oct 13Last article 59m ago ·

Overview

AISummary of 1 article

TokenSpeed, a project from lightseekorg, reports that Qwen3.8-Flash-Next reached 685 output tokens/s per user in a benchmark on four Nvidia GB300 GPUs using NVFP4.

Alibaba's Qwen account congratulated lightseekorg on the result and said it hopes the community will unlock more of the model's inference potential. The figure comes from lightseekorg's benchmark setup as reported in Qwen's post, and no independent measurement is cited.

Written by AI from the articles below · updated Oct 13, 11:45 PM ET

Check the sources:

Latest developmentOct 13, 11:43 PM ET
Qwen3.8-Flash-Next hits 685 output tokens/s per user on Nvidia GPUs

Article timeline

The articles in this story. Times are ET.

Oct 13
  1. QwenOfficial
    Qwen3.8-Flash-Next hits 685 output tokens/s per user on Nvidia GPUs

    AIQwen congratulates lightseekorg on TokenSpeed results showing Qwen3.8-Flash-Next reaching 685 output tokens/s per user. The benchmark ran TokenSpeed on four Nvidia GB300 GPUs using NVFP4. Qwen says it hopes the community will unlock more of the model's inference potential.

Heat trend

Not enough continuous observations to show a trend yet.