Open source Qwen 3.8 arrives

Carlos MartinezAug 15

Deepseek V4 vs Qwen 3.8

An interesting side by side that tells you the strengths of both models. Qwen (vision) looks impressive at a glance but took many iterations and kept reopening to check its work. Deepseek one-shot the code blind (no vision) Looking closer: only

Passing By PixelsAug 15

Okay, I am fairly confident in my hypothesis now.

The secret sauce behind Qwen 3.8 27B becomes almost immediately evident during testing. It is not the training data. In fact, I doubt any SFT was involved at all. The model was simply allowed GRPO with a more liberal reasoning

AstraiaAug 1551

Qwen 3.8 27B in full BF16 on one 96GB M3 Ultra Mac Studio.

54.74GB weights • 58.09GB peak 21.51 tok/s with native MTP speculative decoding—63% faster than no drafter in my test. The video uses real timestamped stream chunks. Exact Hugging Face recipe ↓

JASON MCNABAug 153