Google's Nano Banana 2 Challenges OpenAI in Image Generation

Launched on February 26, the Google Nano Banana 2 model has yet to surpass its iconic predecessor. Although it made a notable entry, it currently ranks second in the Text-to-Image Arena leaderboard, orchestrated by Artificial Analysis. This ranking evaluates model performance in real-time based on user votes, and it also places Nano Banana Pro on the podium.

The Leaders of the Ranking in March 2026

Nano Banana Pro 2, recognized for its adjustable reasoning capabilities and realistic visual production, is the only recent model to break into a ranking dominated by five major players. OpenAI, despite increased competition, maintains its leading position with GPT Image 1.5, launched in December, which continues to stand out in image generation.

Google, OpenAI's main rival in this field, manages to place three of its models in the top half of the ranking. Black Forest Labs closely follows with its various versions of FLUX.2 occupying the fourth, sixth, and seventh positions. ByteDance, with Seedream 4.0 and 4.5, and xAI with Grok Imagine, complete the top 10. Midjourney, on the other hand, has failed to get its alpha v8 version into the ranking, while its v7 remains at a modest 54th place.

Here are the 10 highest-performing image generation models in March 2026:

• GPT Image 1.5 high (OpenAI): 1,265 (Elo score)
• Nano Banana 2 (Google): 1,258
• Nano Banana Pro (Google): 1,214
• FLUX.2 [max] (Black Forest Labs): 1,200
• Seedream 4.0 (ByteDance): 1,185
• FLUX.2 [pro] (Black Forest Labs): 1,181
• FLUX.2 [flex] (Black Forest Labs): 1,180
• Seedream 4.5 (ByteDance): 1,171
• grok-imagine-image (xAI): 1,171
• Imagen 4 Ultra (Google): 1,169

The Ranking Process by Artificial Analysis

To ensure an impartial evaluation of image generation technologies, Artificial Analysis relies on user participation. On its platform, users compare the results of two anonymized models submitted to the same prompt. The results of these confrontations allow for the assignment of an Elo score to each model, which evolves continuously. A lower-ranked model that beats a higher-ranked opponent gains points, while it loses points if it is defeated by a supposedly less capable model.