125B Qwen 3.8 Flash Next reportedly runs on an RTX 4090 at 100T/s

Open Source4 October 20261 min read

In brief

A developer claims to have run the 125-billion-parameter Qwen 3.8 Flash Next model on a single RTX 4090 at 100 tokens per second. The project, shared on GitHub as Strata, drew significant attention on Hacker News. Technical details remain unclear.

125B-parameter Qwen 3.8 Flash Next model
Claimed to run on a single RTX 4090
100 tokens per second inference speed
Open project on GitHub named Strata

What happened

A project shared on GitHub under the name Strata claims that the 125-billion-parameter Qwen 3.8 Flash Next model can run on a single RTX 4090 GPU at 100 tokens per second (100T/s). The story rose to prominence on Hacker News with 376 points.

Why it stands out

Models of this scale normally require multiple high-end data center GPUs. Achieving this speed on a single consumer-grade card has reignited debate about running AI locally.

Open questions

The source material consists only of a headline; technical details such as quantization, memory management, or which inference engine was used were not shared. As a result, the claim has not yet been independently verified.

Why it matters

For readers interested in local AI, running large models on consumer hardware is a significant threshold; however, it's worth being cautious until the claim is independently verified.

Sources

Hacker News (öne çıkanlar)First reported byPrimary source
Run Qwen 3.8 Flash Next (125B) on consumer hardware (RTX 4090) at 100T/s

NVIDIA #open source #local AI #hardware #language model