腾讯混元Image 3.0超越谷歌，登顶AI榜单

Tencent’s Hunyuan Image 3.0 Surpasses Google in AI Race腾讯混元Image 3.0超越谷歌，登顶AI榜单

Tencent’s Hunyuan Image 3.0 has overtaken Google DeepMind’s Nano Banana to become the top-ranked image-generation model on LMArena, a benchmark platform from UC Berkeley researchers. Released in late September, the open-source model features 80 billion parameters, the largest of its kind, claiming performance on par with leading closed-source systems. Trained on 5 billion image-text pairs and 6 trillion text tokens, it showcases Tencent’s growing strength in generative AI. Despite its high capability, the model costs US$0.10 per megapixel to use—significantly higher than Google’s and ByteDance’s offerings—reflecting its heavy computational demand and premium positioning in the global AI landscape.腾讯发布的混元Image 3.0在LMArena排行榜上超越谷歌DeepMind的Nano Banana，成为全球领先的图像生成模型。该模型于9月底推出，拥有800亿参数，是目前最大的开源图像生成模型，性能媲美行业旗舰闭源系统。它基于50亿图文对和6万亿文本数据训练而成，展示了腾讯在生成式AI领域的实力。尽管功能强大，使用成本为每百万像素0.1美元，高于谷歌和字节的同类产品，显示出其高计算需求与高端定位。

腾讯混元Image 3.0超越谷歌，登顶AI榜单