Two Qwen 3.8 27B agents, a designer and a coder, went back and forth more than 50 times on one desktop before calling the project done.
128GB of memory, nothing metered. That round count is what got me. You'd never let a loop run that long against a paid API. You'd cap it at
Testing Qwen 3.8 Flash Next Q4 on the DGX Spark: Day 34.
Asked it to create Pac-Man, and after 24 minutes, it failed. DeepSeek is the only model that's come close. Interesting to see how future versions progress!
WTF guys, why is literally nobody yapping about qwen 3.8 flash rn?!
Tested the exact same one-shot flappy bird prompt, and the speed-cost-output balance is actually insane solid 8/10. yes some barrier placement is kinda weird, and GLM-5.3 flash has slightly smoother visuals,
Your phone got power with superintelligent
Locally Qwen 3.8 27B on a 3090, over Tailscale. agent designing a camera app with Live preview, capture, filters, zoom, vision for object detection, 155 tok/s prompt, 51 tok/s generate,
You have reached the end of the archive
All of qwen38