AI HOT (Curated Pool)Jul 22, 2026, 20:34 UTC+8

Qwen-Image-3.0 tested: solid Chinese text rendering and multi-image fusion, rivals GPT Image2

实测Qwen-Image-3.0：19大场景挑战GPT Image2，中文长文本与多图融合表现亮眼

Alibaba's Qwen-Image-3.0 is live with 4.5k token input, 12 languages, and 20+ fonts. Across 19 scenarios—Chinese long text, multilingual layout, UI design, multi-image fusion—text stays clean and info aligns correctly. Image quality is comparable to GPT Image2. Available in China with fast speed and affordable pricing. The post doesn't disclose exact pricing or latency figures.

Why it matters: A 19-scenario hands-on test of Alibaba's Qwen-Image-3.0, with clear strengths in Chinese long-text rendering and multi-image fusion. The GPT Image2 comparison is convincing. Docked a few points because the post doesn't disclose specific pricing or latency numbers, and it's a s...

Read the original ↗Export Markdown

Take

This one's worth opening because Alibaba finally tackled the hard part: Chinese long-text image generation that doesn't fall apart. 4.5k token input, 12 languages, 20+ fonts—and across 19 test scenarios, the text stays clean and info aligns correctly. That's rare for a domestic model. Quality is comparable to GPT Image2, and it's usable in China right now.

I'd discount this a bit: the post is an RSS snippet with no pricing or latency numbers, just "affordable" and "fast." If you're putting this into production, those two numbers matter more than 19 demo scenarios. Also, GPT Image2 isn't the absolute benchmark for image gen—"comparable" is more of a usability signal than a win.

What's missing: a pricing table, API latency, concurrency limits, and side-by-side comparisons with Flux or SD3. If the price is genuinely low, stable Chinese typography alone has real value—posters, covers, e-commerce images all demand text accuracy.

Source: AI HOT (Curated Pool) · mp.weixin.qq.com
Published: Jul 22, 2026, 20:34 UTC+8 · Jul 22, 2026