Zhipu and Huawei just open-sourced GLM-Image — and the real shock is that it was trained end-to-end on China’s domestic AI stack, using Huawei Ascend Atlas 800T A2 hardware and MindSpore. That’s a straight signal that China is building an alternative compute ecosystem that can compete at scale. Then Google upgrades Veo 3.1 with reference-image video generation, native vertical output, 1080p plus 4K upscaling, and SynthID watermarking, rolling it out across Gemini, YouTube Create, and Vertex AI like it’s about to flood the entire internet with controllable AI video. After that, Google pushes into the real world with MedGemma-1.5 and MedASR, built for CT, MRI, pathology slides, lab report extraction, and clinical speech-to-text — basically turning medical AI into something that looks workflow-ready. And finally, Rokid unveils AI glasses that integrate ChatGPT, record 4K, translate in 89 languages, and even add payments through Alipay+ GlassPay, making AI wearables feel like an actual consumer platform shift.
智谱与华为刚刚开源了GLM-Image模型——真正令人震动的是,它完全基于中国本土AI技术栈进行端到端训练,采用华为昇腾Atlas 800T A2硬件与MindSpore框架。这清晰表明中国正在构建能够规模化竞争的自主计算生态。随后谷歌升级了Veo 3.1视频生成模型,新增参考图像生成视频、原生竖屏输出、1080P及4K超清升级、SynthID数字水印等功能,通过Gemini、YouTube Create和Vertex AI平台铺开,其势仿佛即将用可控AI视频席卷整个互联网。紧接着,谷歌又通过MedGemma-1.5与MedASR切入现实医疗场景,专为CT、MRI、病理切片分析、实验室报告提取及临床语音转文本打造——实质上将医疗AI推向了工作流就绪阶段。最后,Rokid发布的AI眼镜集成了ChatGPT、支持4K录制、89种语言实时翻译,甚至通过Alipay+ GlassPay实现支付功能,让AI可穿戴设备展现出真正的消费级平台变革态势。