Alibaba
Alibaba Crams Large Models Into 3090 GPUs — Local AI Is No Longer a Geek Toy
Qwen community leaks 3.8 Flash Next: same architecture as Qwen 4, single 3090 GPU hits near-cloud speed. Not a tech upgrade — a compute threshold drop
Back to home
on-device inference
3 articles tagged with this topic
Tongyi Qianwen
27B Is Already Enough — Tongyi's New Model Ends the 'Bigger Is Better' Story
On AMD Strix, Tongyi's newest model barely beats 27B in daily work — a 'good enough' signal more valuable than any benchmark for enterprise IT.
world model
有人在 iPad 上跑通了「会看 照片的游戏引擎」——AI 生成游 戏画面的实验已经小到装 进口袋
A developer trained and ran a world model on iPad locally, turning any photo into a drivable game scene — no cloud required.