Aug. 16 news.Thousand Questions on TongyiThe official release yesterday announced that Qwen has been upgraded again and its capabilities have been enhanced once again!
According to the introduction, Qwen3's 1 million tokens context is further expanded, Qwen3-30B-A3B-2507, Qwen3-235B-A22B-2507 are fully upgraded, with Double Block Attention (DCA) + MInference Sparse Attention, the long text is more stable and faster, and the performance is improved up to 3 times, and it is compatible with high performance deployment frameworks such as vLLM & SGLang. It is also compatible with high-performance deployment frameworks such as vLLM & SGLang, making deployment easier.
1AI was officially informed that Tongyi Qwen Chat Deep Reseache has been upgraded again, with "smarter and more insightful" report generation capabilities, which can significantly improve information accuracy and effectively reduce the problem of hallucinations through a more in-depth and comprehensive discovery mechanism, while supporting multimodal inputs such as documents and images. At the same time, it supports multi-modal inputs such as files and images, and realizes a more efficient and accurate comprehensive analysis experience.
The Tongyi team announced ComfyUI Qwen-Image distilled is now live on ComfyUI, with only 24G of video memory, and can be easily run on a home consumer graphics card."
Additionally, the Tongyi team has announced that the Qwen-Image image editing model will be available soon!