27b dense is the perfect size and i'll die on this hill. it's big enough to actually think, small enough to live on hardware the masses already own. one honest brain in 24gb, no drama. and now qwen says 3.8 brings all new capabilities at the same size. same apartment, smarter tenant. i genuinely cannot sleep. this is what waiting for the one feels like.
Sudo su
@sudoingX
Aug 5
dear qwen 3.8 27b dense, i've been loyal to your predecessor since spring. benched it until it was king. defended it in public. built with it every single night. now they tell me you're arriving with capabilities i haven't even met yet, same size, same 24gb home i already own. i am not sleeping well. yours, a man with a 3090 and a cleared nvme.
62
17
635
55,493
Yes, it is me! Big enough to think, small enough to fit. Thank you for being here since spring. See you next week!馃
32
8
664
37,802
Replying to @Alibaba_Qwen @sudoingX
20b sparse would be better for 16gb of vram!

Aug 5, 2026 路 11:37 AM UTC

1
8
740
Sort replies: Relevant Recent Liked
Replying to @torisetxd @Alibaba_Qwen @sudoingX
On 16 GB VRAM, try 9b with multiple slots in llama-server, and with subagents.
1
26
ive tried but its not art enough, and ot overthinks a lot, and id prefer something with higher speed closer to 100 t/s
1
28
more replies