Summary and commentary: Geek Haus. Editorial responsibility: Jay Jang, CEO
Back to feedStrata claims Qwen3.8-Flash-Next 125B can run locally on consumer GPUs like RTX 4090 at about 100 tokens per second
Geek Haus editorial team · Sources and editorial policy
EDITOR BRIEF
The Strata project says its open-source software can run the Qwen3.8-Flash-Next 125B-parameter model on ordinary Windows or Linux gaming PCs with NVIDIA or AMD GPUs. It emphasizes local inference for chat, coding, image understanding, and agent workflows, with no data leaving the user’s machine.
COMMENTARY
If the performance claims hold up, this points to rapid progress in making very large models usable outside cloud infrastructure. The trend could strengthen demand for high-memory consumer GPUs and accelerate interest in local AI for privacy-sensitive coding, productivity, and automation use cases.
COMMENTS
Discussion
> geekhaus:~$ read next
Related articles
dgt.is
Developer argues DeepSeek 4.1 Flash delivers frontier-like AI performance at dramatically lower cost than OpenAI and Anthropic models
research.meta.ai
Meta releases Muse Spark Open, an 816B-parameter open AI model tuned to run locally on Mac Studio
Cloudflare Blog