Tencent Launches an Image Model on Alibaba's Big Day
Hy Image 3.5 Preview went live on Tuesday, timed to Alibaba's Apsara conference. Tencent says in-house designer trials put it level with ByteDance's Seedream 5.0 Pro, but published no hard numbers.
Tencent released Hy Image 3.5 Preview on Tuesday, and the timing was not accidental: it landed the same day Alibaba held its annual Apsara conference in Hangzhou. The model does text-to-image and image-to-image in one system, which means you can generate a picture and then keep editing it in the same conversation rather than starting again. It accepts up to five reference images in a single request, handles the usual range of shapes, and tops out at 2K resolution, priced at 0.15 RMB per 2K image, a few cents. It is rolling into the Yuanbao chatbot along with Tencent’s film-editing and design tools, and is available through Tencent Cloud. Shares rose more than 7 percent in Hong Kong.
The performance claim needs a careful reading. Tencent says trials with hundreds of its own in-house designers put the outputs on par with ByteDance’s Seedream 5.0 Pro and slightly ahead of Google’s Nano Banana Pro and Alibaba’s Qwen-Image-3.0 Pro. It published no hard metrics to go with that. Preference tests run by the company that made the model, judged by staff of the company that made the model, are the weakest form of evidence in this field, and everyone in it does them.
What is behind this
The competitive picture in Chinese image generation is now four large companies shipping models within weeks of each other, all claiming rough parity, all pricing at a few cents per image. That tells you something the benchmark claims do not: this capability has commoditised. When four rivals produce comparable results at comparable prices, the product stops being the model and starts being where it is plugged in, which is why the more interesting line in Tencent’s announcement is the list of places it appears on day one rather than any quality claim. Tencent owns WeChat, which is where a billion people already are.
Merging generation and editing into one model is a genuine usability improvement rather than a benchmark one. The older split, where one model made pictures and another modified them, meant every edit was a fresh negotiation with a system that had no memory of what it had just made. Keeping the conversation intact is the difference between a slot machine and a tool.
What this means for you: Outside China this is not something you will use, and Yuanbao is not a realistic option for most readers here. The reason to note it is price and direction. Image generation at a few cents a picture, good enough that four companies argue about single-digit differences, is now a commodity input rather than a special capability, and that is the context for any product you see charging a premium for it. If you already use an image tool you like, there is no reason to switch today.
Sources
Anthropic Has a Real Laboratory Now, With Pipettes and Everything
Anthropic's head of life sciences confirmed to Reuters that the company runs a Bay Area wet lab where Claude's biological hypotheses get tested physically. The company says it is not becoming a drug maker.
