Qwen 3.8 27B open model rivals frontier, tops Hugging Face

Alibaba's Apache-2.0 Qwen 3.8 27B scores 52 on the Artificial Analysis Intelligence Index, matching GPT-5.6 Luna (max) and one point behind GLM-5.2 and DeepSeek V4 Pro. It runs locally on an RTX 5090 at up to 200 tk/s, but defaults to 'xhigh' reasoning effort causing overthinking.
How this story unfolded
12 days · 4 reports · 41 community posts · from Aug 14
- Aug 14
- Aug 15
- Aug 16
- Aug 17
- Aug 18
- Aug 19
- Aug 22
- Aug 23
- Aug 24
- Aug 25
- Aug 26
Qwen by email
Get an email when Qwen has news
No news that day, no email.
More stories today
- Stable Diffusion user tests H3 model with Cheers-style script
- Reddit users share impressive image-to-video AI demos
- Reddit reminds users they can legally seed AI models via torrenting
- MiniMax H3 reverse-engineers paintings into basic forms
- OpenAI DevDay Exchange Seoul applications close Sept 4