Google’s Gemini Models Sweep Top Spots in December 2025 Text-to-Image Arena Rankings
The December 2025 Text-to-Image Arena leaderboard has revealed a dramatic reshuffling of AI image generation models, with Google’s Gemini family capturing the top two positions and establishing a commanding lead over the competition. Updated on December 3 with over 3.6 million total votes across 34 models, the rankings demonstrate Google’s accelerating dominance in the text-to-image space.
Google’s Commanding Performance
Google’s new Gemini 3 Pro image models have secured the #1 and #2 positions with scores that significantly outpace the competition. The gemini-3-pro-image-preview-2k (nano-banana-pro) leads the pack with a score of 1234 (±10 confidence interval), while its sibling model gemini-3-pro-image-preview follows closely at 1231 (±7). This 70+ point gap between Google’s top models and the third-place contender represents the largest performance margin observed in recent leaderboard updates.
Google’s success extends beyond the flagship Gemini 3 Pro models. The gemini-2.5-flash-image-preview holds the #4 position with a score of 1156, while imagen-4.0-ultra-generate-preview and imagen-4.0-generate-preview occupy #8 and #10 respectively. With four models in the top ten, Google has achieved unprecedented coverage across multiple performance tiers.
The Challenger Landscape
While Google dominates the summit, several competitors are mounting credible challenges.
Black Forest Labs’ Flux continues to prove itself as the leading open-source alternative. Flux-2-flex claims the #3 spot with a score of 1157, essentially tied with Google’s Gemini 2.5 Flash model, while Flux-2-pro secures #6 at 1146. The Flux family demonstrates remarkable consistency, with five total variants ranked throughout the top 30.
Chinese tech giants have emerged as serious contenders. Tencent’s hunyuan-image-3.0 ranks #5 with 1154, while Bytedance’s Seedream family occupies three positions in the top 11, including the newly-previewed seedream-4.5 at #7 (1146) and seedream-4-2k at #9 (1144). This concentration of Chinese-developed models in the top tier signals a significant shift in the global competitive balance.
Notable Rankings and Trends
The leaderboard reveals several surprising developments:
- OpenAI’s gpt-image-1 sits at #13 with a score of 1119, trailing well behind Google’s top performers and failing to crack the top ten
- DALL-E 3 has fallen to #30 (978), marking a dramatic decline for the model that helped pioneer the space
- Microsoft’s mai-image-1 debuts at #16 with 1091, positioning itself as a mid-tier contender
- Alibaba’s wan2.5-t2i-preview ranks #12 (1123), while their qwen-image models occupy #19 and #22
- Stable Diffusion 3.5 Large sits at #33 (940), continuing Stability AI’s struggle to compete at the highest level
Key Takeaways
- Google has established a two-tier lead with Gemini 3 Pro models scoring 70+ points ahead of the competition
- The performance gap between proprietary and open-source models is widening, though Flux maintains a strong third-place position
- Chinese AI labs are consolidating power with Tencent and Bytedance securing multiple top-tier rankings
- OpenAI’s image generation capabilities are stagnating relative to competitors, with both gpt-image-1 and DALL-E 3 underperforming
- The total vote count of 3.6 million indicates robust community engagement and statistical reliability
Verdict: A New Hierarchy Emerges
The December 2025 leaderboard marks a potential inflection point for the AI image generation market. Google’s simultaneous launch of multiple best-in-class models suggests a maturation of their multimodal strategy, while the strong showing from Chinese labs indicates that the competitive landscape is becoming increasingly multipolar.
For practitioners and enterprises, the message is clear: Google’s Gemini family currently represents the performance ceiling for text-to-image generation, but viable alternatives exist for organizations seeking open-source solutions (Flux) or requiring region-specific optimizations (Tencent, Bytedance). The dramatic decline of previously dominant models like DALL-E 3 serves as a reminder that leadership in this space remains fluid and highly competitive.
As we enter 2026, all eyes will be on whether competitors can close the substantial gap Google has opened—or if the search giant will continue to extend its lead in the next generation of image generation models. For those looking to explore the current landscape of AI image generators, our complete guide to the best AI image generators in 2025 provides detailed comparisons and recommendations.
The rise of Google’s Gemini models also highlights the importance of understanding how different AI image generators compare in real-world applications. As the technology continues to evolve rapidly, staying informed about the latest developments in AI image generation news becomes crucial for professionals and enthusiasts alike.