Arena: GPT Image 2 Tops August 2026 Rankings

Le brief IA que les pros lisent chaque soir
Les 7 actus IA du jour, décryptées en 5 min. Gratuit.
Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.
Choisis ton rythme
Gratuit · Pas de spam · Désabonnement en 1 clic
Two distinct tables now structure the performance of image generators: text-to-image and editing. GPT Image 2 leads both as of August 2026, ahead of Microsoft's MAI-Image-2.6 preview. However, the discrepancies in vote volume and the "preliminary" label for less contested models leave room for movement. Meta makes a notable entry with Muse Image, while Google and OpenAI each only have one model in the top 10 generation.
Arena Displays Uncertainty Intervals
The editing table aggregates over 29 million votes, compared to nearly 5.9 million for text-to-image generation, covering 53 models. Arena breaks down editing into seven categories, from photorealism to text rendering, also separating multi-image editing. The platform publishes a confidence interval and a range of possible positions for each model to reflect the uncertainty related to the number of votes. Systems that are too little contested receive the "preliminary" label. In this context, the MAI-Image-2.6 preview accumulates 3,488 votes, and Grok Imagine 2.0 has 2,676, the latter carrying the "preliminary" tag. Their rankings are therefore likely to change.
GPT Image 2 Remains First in Editing as of August 2026
Arena rethought its tables this summer and now distinguishes the ability to generate an image from the ability to edit an existing visual based on a prompt. In editing, GPT Image 2 holds the top position as of August 2026. Grok Imagine 2.0 ranks second, Muse Image fourth, and MAI-Image-2.6 preview fifth, with respective Elo scores of 1,463, 1,439, 1,406, and 1,420. MAI-Image-2.5 follows with 1,401. Google appears more comfortable in this exercise with Nano Banana Pro in eighth position, while Reve slips to twelfth place.
The Top 10 Text-to-Image and Their Elo Scores
The text-to-image ranking for August 2026 places GPT Image 2 medium (OpenAI) at the top with an Elo score of 1,381, followed by MAI-Image-2.6 preview (Microsoft) at 1,336 and Grok Imagine 2.0 low (SpaceXAI) at 1,316. Following them are Reve 2.1 (1,302), Muse Image (1,282), and Reve 2.0 (1,270). The remainder of the top 10 includes Google’s Nano Banana 2 (1,264), ByteDance’s Seedream 5.0 Pro (1,258), Alibaba’s Qwen Image 3.0 Pro (1,257), and Microsoft’s MAI-Image-2.5 (1,256).
Movements of Players in August
Muse Image, the first model developed in-house by Meta and launched on Instagram and WhatsApp at the beginning of July, ranks fifth in generation. Reve retains two models in the top 10, occupying the fourth and sixth positions, despite the departure of its research team to OpenAI at the end of July. On Google’s side, Nano Banana 2 appears as the only model from the group to feature in the top 10, in seventh place, while the two versions of Nano Banana Pro occupy the twelfth and fourteenth positions, respectively. OpenAI does not include a second model in the top 10, with GPT Image 1.5 landing in thirteenth place. Alibaba completes this table with Qwen Image 3.0 Pro in ninth position.
Brief IA — L'actualité IA en français
L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.