@estop845: alibaba just announced qwen 3.8 and the open weights drop next week. the flagship trades blows with opus 4.8 and fable 5 — wins vision, perception, and computer use, loses some coding. but the one that actually matters is the small one. qwen 3.8 27B runs locally on 17gb of ram. that’s a box on your desk, not a cloud bill. nothing leaves your machine, no per-token cost, no usage caps, nobody deciding next quarter that your plan costs more now. if you’re building anything with client data — legal, medical, financial — that changes everything. i’m moving my agent workloads local the week these weights drop. #qwen #opensourceai #ai #aiagents #aiforbusiness
estop
Region: US
Monday 03 August 2026 19:21:02 GMT
Music
Download
Comments
... :
I'm hoping 35B doesn't take too long to come out as 27B is so slow when you have to offload to cpu
2026-08-04 06:53:05
2
Voxaris :
if i were hiring an ai company a huge dealbreaker would be this. it’s super important to say you own the data. it’s not sourced through a 3rd party. that’s a big liability!
2026-08-03 19:41:37
12
riyuuk :
17GB VRAM? That’s close to a 5090 which is 4000$
2026-08-03 23:32:04
2
Professional Street Racist :
How many t/s tho if its ram
2026-08-03 23:08:17
2
🅲🆄🅽🆃 :
I love quen, 32b is a little too big, and 14b does me perfect on 2.5, looking forward to this
2026-08-04 02:10:56
2
Rhett.Richter :
🤯🤯🤯
2026-08-04 00:29:14
1
To see more videos from user @estop845, please go to the Tikwm
homepage.