Alibaba releases Qwen3.8-Flash multimodal MoE, preview of Qwen4 architecture
Qwen3.8-Flash is a 125B-parameter multimodal MoE with 51B active parameters, 262K native context (extensible to 1M via YaRN), and open weights. QwenCloud API pricing is $0.16/1M input and $0.47/1M output tokens. It's now available on Vercel's AI Gateway.
How this story unfolded
1 day · 1 report · 5 community posts · from Aug 25
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