Brother ... dream big. 7900xtx running qwen 3.8 27B , pretty smooth.
Ignore the bumbled video in the final few minutes( for some weird reason my gnome screen camera got bugged) but as you can see, the model works fine.
This had to happen and it did.
2 most popular Moe contenders: 🔵 Ornith 1.5 35B-A3B (4-bit) 🟣 Qwen 3.6 35B-A3B (4-bit) I can already tell you that if Qwen would decide to give us a 3.8 35B A3B this would be the best 3b active parameter moe on the market and could actually
Qwen 3.8 Flash Next has 125B parameters but only activates 6B per token with a 1M token context window
Scoring 84.5 on AndroidWorld and 62.5 on SWE-Bench Pro for real-world coding.
DeepSeek-V4-Flash-Vision, Qwen 3.8 Max, Claude credits, and agent compute - all FREE right now.
HuggingFace: DeepSeek-V4-Flash-Vision-Exp, 305B, first multimodal DeepSeek, MIT open weights: Qwen Chat: Qwen 3.8 Max free, 1M context, image input, zero
You have reached the end of the archive
All of qwen38