Qwen 3.8 27B built my city at night at 65 tok/s.
Every window was piled at the origin. Same prompt on three stacks. DeepSeek V4.1 Flash on the 4 box Spark ring hit 120.74 tok/s. GLM 5.3 Flash on TensorFold 0.6.0 TP4 hit 108.0 tok/s. Qwen 3.8 27B on the Mac Studio alone hit 65.23
Man, would you be so kind to run the Burj Al Arab test again?
Reaaaaally would like to compare it to my Qwen 3.8 Flash result. This one was created with Unsloth's Q6_K_XL using the DeepSeek Harness.
We have live Model APIs at urunml.
Here's Qwen 3.8 27B running on uRun, plugged into OpenCode over industry-standard chat completions. Same requests your code already sends, open model underneath. Read more here:
Model APIs are live on uRun.
In the demo, we plug Qwen 3.8 27B into OpenCode and chat with it over industry-standard chat completions. Working internally, rolling out worldwide. Our waitlist gets access first: More:
You have reached the end of the archive
All of qwen38