Included UD-Q3 and UD-IQ4 in my Qwen 3.8 27B quant benchmark.
Feed the model a task as an image card, then see if it can produce the animation. I can't say my version is 'better' exactly, but I will say that from every test I've run and 2 individual reviewers, it's near
Qwen 3.8 at 51tps on Studio M5 Ultra via MLX-VLM ordering me sushi for dinner via computer use.
(Yes I know there is a DoorDash cli - this is an eval I’m called SushiBench)
SITUATION EXPLAINED: Alibaba published its AI roadmap.
Qwen 5 will be 5 to 10 trillion parameters. • Qwen 4 is in training now, with Qwen 4.5 and Qwen 5 scaling to 5 to 10 trillion, roughly the size of Mythos, Fable, and Astra • Qwen 3.8 is around 2.4 trillion, so that's a 2
I had Opus 5.5 build a tiny Sims town where every resident's dialogue is written live by a local AI 🏘️🧠
Then I became the sky ☁️ and dropped a tornado on them 🌪️ and abducted the mayor with a UFO 🛸👽 Now they gossip about it 😂 Running 100% local on Qwen 3.8 27B + my 3090s
You have reached the end of the archive
All of qwen38