Qwen4-Max-Preview Launches with Strong Coding and 3D Tests
Qwen4-Max-Preview has launched as a preview model, and PingWest reports that it improved coding, 3D modeling, and interactive web creation in hands-on tests. The model handled website changes accurately and produced clearer internal structures than Kimi K3.1 Pro in selected Blender and 3D web tasks, while Kimi K3.1 Pro showed stronger aesthetics. Qwen did not disclose parameter scale, benchmark scores, training data, or detailed architecture changes, and comparisons with Fable5.2 and GPT-6 Astra remained limited to the reported demos.
The preview's launch and practical tests show it can handle coding, website modification, 3D modeling, and interactive web tasks, but the undisclosed specifications and limited comparisons leave its position against frontier models uncertain.
Next Week 了半天,Qwen4-Max-Preview 终于落地了。随着预览版正式上线,一个悬念也跟着揭晓。前几天,Qoder 放出一个叫「Qoduck」的匿名模型,其实就是 Qwen4-Max-Preview。
Qwen 没有披露具体参数规模、基准测试成绩和训练数据。只在官方公告中提到,这一代补强了底层的世界知识与多模态感知,也提升了 Coding 和 Agentic 能力。
结合此前的报道来看, Qwen4 这次对底层模型架构进行改动,这套新架构究竟带来什么变化,可能还要等更多的信息。
#Qwen4-Max-Preview#Qwen#Qoder#Qoduck#Kimi K3.1 Pro#Fable5.2#GPT-6 Astra