Recraft continues to dominate the Text to Image Arena, a ranking that objectively evaluates the performance of AI models for image generation.
“Understanding the AI landscape to identify the model and provider best suited to your needs” : this is the ambition of the Text to Image Arena, a ranking that evaluates, in real-time, the performance of image generation models available on the market. Inspired by the principle of the Chatbot Arena and managed by the independent organization Artificial Analysis, the Text to Image Arena positions itself as an alternative to companies’ internal evaluation methods, sometimes criticized for their lack of transparency.
To ensure an impartial assessment, Artificial Analysis relies on user participation: on the platform, users are invited to compare the results of two anonymized models generated from the same prompt. So, which technologies will dominate the market at the start of 2025? Below are some answers.
The 10 Most Powerful AI Image Generators in January 2025
A hierarchy is emerging in the image generation model market: this month, the positions remain fixed. In January 2025, the Recraft V3 model, initially hidden under the identity red_panda, retained the first position, from which it has not been dislodged since October 2024. This technology, which powers the eponymous platform, allowing the generation of images in more than 100 different styles, maintains a slight lead over FLUX 1.1 [pro] and FLUX.1 [pro], developed by Black Forest Labs.
With version 6.1, released in July 2024, Midjourney remains at the foot of the podium, while version 6 is in 7th position. Furthermore, Ideogram and Stability.ai remain well-established in the first half of the table. Ideogram’s technologies are positioned in 6th and 8th place, while Stability.ai’s technologies retain 9th and 10th positions. Adobe, OpenAI and Luma Labs are still significantly behind their competitors.
Here are the 10 best-performing image-generation models as of January 2025:
Creator | Name | Arena ELO | Appearances |
Reve | Reve Image (Halfmoon) | 1143 | 6305 |
Recraft AI | Recraft V3 | 1141 | 112843 |
Imagen 3 (v002) | 1115 | 20164 | |
Black Forest Labs | FLUX1.1 [pro] | 1113 | 130518 |
Black Forest Labs | FLUX.1 [pro] | 1098 | 140704 |
Midjourney | Midjourney v6.1 | 1079 | 138752 |
MiniMax | Image-01 | 1079 | 6731 |
Black Forest Labs | FLUX.1 [dev] | 1076 | 139101 |
Ideogram | Ideogram v2 | 1075 | 138928 |
Midjourney | Midjourney v6 | 1073 | 164498 |
Ideogram | Ideogram v2 Turbo | 1067 | 139681 |
Stability.ai | Stable Diffusion 3.5 Large Turbo | 1065 | 125002 |
Luma Labs | Luma Photon | 1062 | 18290 |
Stability.ai | Stable Diffusion 3.5 Large | 1061 | 125461 |
Bytedance | Infinity 8B | 1055 | 7053 |
Ideogram | Ideogram v1 | 1055 | 141595 |
Leonardo.Ai | Phoenix 1.0 Ultra | 1052 | 8943 |
Stability.ai | Stable Diffusion 3 Large | 1048 | 159440 |
Black Forest Labs | FLUX.1 [schnell] | 1035 | 142057 |
Playground AI | Playground v3 (beta) | 1032 | 143050 |
Ideogram | Ideogram v2a | 1026 | 7372 |
Leonardo.Ai | Phoenix 0.9 Ultra | 1025 | 94813 |
Leonardo.Ai | Phoenix 1.0 Fast | 1015 | 9550 |
Ideogram | Ideogram v2a Turbo | 1012 | 7640 |
Recraft AI | Recraft 20B | 1011 | 135709 |
Luma Labs | Luma Photon Flash | 998 | 28187 |
Gemini 2.0 Flash Experimental | 994 | 7943 | |
Playground AI | Playground v2.5 | 990 | 163397 |
OpenGVLab | Lumina Image v2 | 988 | 14810 |
Adobe | Adobe Firefly 3 | 976 | 122843 |
OpenAI | DALLE 3 HD | 975 | 164415 |
Stability.ai | Stable Diffusion 3.5 Medium | 968 | 8594 |
OpenAI | DALLE 3 | 960 | 164699 |
Stability.ai | Stable Diffusion 3 Medium | 937 | 151257 |
Amazon | Amazon Titan G1 v2 (Standard) | 933 | 150382 |
Stability.ai | Stable Diffusion 3 Large Turbo | 931 | 160516 |
Stability.ai | Stable Diffusion 1.6 | 919 | 159518 |
Amazon | Amazon Titan G1 (Standard) | 915 | 153970 |
Stability.ai | SDXL Lightning | 908 | 163992 |
Stability.ai | Stable Diffusion XL 1.0 | 883 | 164665 |
OpenAI | DALLE 2 | 748 | 211142 |
Stability.ai | Stable Diffusion 2.1 | 746 | 211291 |
DeepSeek | Janus Pro | 729 | 25267 |
Stability.ai | Stable Diffusion 1.5 | 658 | 143776 |
What are the ranking criteria for the Text to Image Arena?
In the Artificial Analysis rankings, each model receives an Elo score, just like in chess competitions or competitive gaming ladders, which changes after each duel organized on the platform. Specifically, if a model wins against a higher-ranked opponent, it earns points. Conversely, its Elo score drops if it loses to a supposedly weaker model. According to Artificial Analysis, the rankings are updated every hour.