Tencent's Hunyuan Image 3.0 Tops AI Leaderboard, Surpassing Google's Nano Banana
- tech360.tv
- 10 hours ago
- 2 min read
Tencent Holdings' new artificial intelligence model, Hunyuan Image 3.0, has surpassed Google DeepMind’s "Nano Banana" to become the leading image-generation model. This includes both open-source and closed products, according to a major public leaderboard.
As of Saturday, the open-source Hunyuan Image 3.0 secured the top position in the text-to-image rankings on LMArena. This AI model evaluation platform was originally started by researchers at the University of California, Berkeley.
Google DeepMind’s Gemini 2.5 Flash Image, also known as "Nano Banana," was the previous leader. It rapidly gained popularity after its release in late Aug. due to its image editing accuracy and 3D-figurine generation.
Tencent released Hunyuan Image 3.0 late last month, stating it was "completely comparable to the industry’s flagship closed-source models." It is the industry’s biggest open-source image-generation model to date.
The model boasts 80 billion parameters, making it the largest open-source image-generation model. Parameters are variables that encode a model’s intelligence and are adjusted during training.
Generally, a higher number of parameters indicates a more powerful model. However, it also requires greater computational resources to train and operate.
On Fal.ai, a global generative media hosting platform, Hunyuan Image 3.0 costs USD 0.10 per megapixel generated. For comparison, Gemini 2.5 Flash Image is priced at USD 0.039 per image.
ByteDance’s flagship model, Seedream 4.0, costs USD 0.03 per image. Tencent said its model was trained on an extensive data set.
This data set comprises 5 billion image-text pairs, video frames, interleaved image-text data, and 6 trillion tokens of text. Tencent stated in its announcement, “Whether you’re an illustrator, designer or creator, this is built to slash your work flow from hours to minutes.”
Hunyuan Image 3.0 is available on Tencent’s official website and has been open-sourced on developer platforms GitHub and Hugging Face. It is trending alongside other recently launched Chinese models, such as DeepSeek’s V3.2-Exp and Z.ai’s GLM-4.6.
LMArena ranks models based on user preferences; users compare two images generated by different models and vote for their favourite. The platform has labelled Hunyuan Image 3.0’s ranking as “preliminary,” meaning current votes are based on a pre-release version of the model.
The model currently supports only text-to-image generation. Tencent plans to introduce additional functionalities, including image-to-image generation and image editing.
Tencent’s Hunyuan Image 3.0 leads the LMArena text-to-image rankings.
The model has 80 billion parameters, making it the largest open-source image-generation model.
It surpassed Google DeepMind’s "Nano Banana" (Gemini 2.5 Flash Image) for the top spot.
Source: SCMP