Tencent’s new AI model, Hunyuan Image 3.0, is now ranked as the top image-generation model on LMArena, an AI evaluation platform started by researchers at the University of California, Berkeley.

LMArena’s leaderboard currently puts Hunyuan Image 3.0 ahead of Google DeepMind’s Gemini 2.5 Flash Image, also known as Nano Banana.

Hunyuan Image 3.0 is open source and has 80 billion parameters, making it the largest open source image-generation model so far.

Tencent trained it on 5 billion image-text pairs and 6 trillion text tokens.

The model currently supports text-to-image generation and costs US$0.10 per megapixel.

🔗 Source: South China Morning Post