---
format: "aidr-story-markdown/v1"
id: "737b6f8c68e96bf9a48fb738561fa4af38e277ce056d0a270cc85057d1403ccb"
canonical_url: "https://aidr.today/737b6f8c?lang=en"
title: "SGLang Nears 1,000 Tokens a Second per User on DeepSeek V4.1 Flash With Nvidia GB300"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-10-29T15:21:10.000Z"
category: "Frameworks"
topics: ["deepseek","inference","framework"]
source_urls: ["https://huggingnews.com/ai/sglang-nears-1000-tokens-a-second-per-user-on-deepseek-v41-flash-with-nv-12da453c","https://x.com/SemiAnalysis_/status/2115821309137465536","https://x.com/sgl_project/status/2114071115605877043","https://x.com/sgl_project/status/2114071117669536251","https://x.com/vllm_project/status/2114054277379641770","https://x.com/bookwormengr/status/2112636749293949122","https://x.com/swyx/status/2112587397380943953","https://x.com/zephyr_z9/status/2112570318686683281"]
summary: "Open source inference framework SGLang reached nearly 1,000 tokens per second per user running DeepSeek-V4.1-Flash on an Nvidia GB300 NVL72 system, according to SemiAnalysis. The performance gain stems from combining prefill and decode disaggregation with kernel fusion techniques, and SGLang achieved up to 1.04 times DeepInfer's prefill throughput on four GB300 or B300 GPUs and up to 7.4% higher decode throughput across 64 GB300 chips."
---

# SGLang Nears 1,000 Tokens a Second per User on DeepSeek V4\.1 Flash With Nvidia GB300

> [Open the canonical story](<https://aidr.today/737b6f8c?lang=en>)

**Published:** 2026-10-29T15:21:10.000Z
**Category:** Frameworks
**Topics:** deepseek, inference, framework

## Summary

Open source inference framework SGLang reached nearly 1,000 tokens per second per user running DeepSeek\-V4\.1\-Flash on an Nvidia GB300 NVL72 system, according to SemiAnalysis\. The performance gain stems from combining prefill and decode disaggregation with kernel fusion techniques, and SGLang achieved up to 1\.04 times DeepInfer's prefill throughput on four GB300 or B300 GPUs and up to 7\.4% higher decode throughput across 64 GB300 chips\.

## Sources

- [Story source](<https://huggingnews.com/ai/sglang-nears-1000-tokens-a-second-per-user-on-deepseek-v41-flash-with-nv-12da453c>)
- [Story source](<https://x.com/SemiAnalysis_/status/2115821309137465536>)
- [Story source](<https://x.com/sgl_project/status/2114071115605877043>)
- [Supporting source](<https://x.com/sgl_project/status/2114071117669536251>)
- [Supporting source](<https://x.com/vllm_project/status/2114054277379641770>)
- [Supporting source](<https://x.com/bookwormengr/status/2112636749293949122>)
- [Supporting source](<https://x.com/swyx/status/2112587397380943953>)
- [Supporting source](<https://x.com/zephyr_z9/status/2112570318686683281>)