OpenAI once again takes the podium in the Text-to-Image Arena, Arena’s ranking that differentiates image generators based on internet users’ votes, for the second consecutive month. GPT Image 2.5 Sunburst remains first, ahead of GPT Image 2.5 Flare and GPT Image 2, as in September. Two models newly added to the chart, from Google and SpaceXAI, immediately settle into the top 6.
- OpenAI leads the Text-to-Image Arena for the second month in a row, with GPT Image 2.5 Sunburst in first place.
- Nano Banana 2.1 and the new Grok Imagine 2.0 enter the top 6 immediately.
- In image editing, the same OpenAI trio leads, while Krea 3 Ultra debuts in eighth place.
- Arena ranks models through blind user votes, and recent entries remain marked as preliminary until more votes arrive.
Image Generation: The Top 10 of October 2026
Launched on October 6, Nano Banana 2.1 is an update of Nano Banana 2, Google’s image model, known in the API as Gemini 3.1 Flash Image. It enters in sixth place, with a still provisional score due to insufficient votes. SpaceXAI places two versions of Grok Imagine 2.0 in the ranking: the most recent, also provisional, climbs to fourth position, while the August 1 version falls to eighth place as its votes accumulate.
Between these two versions, MAI-Image-2.6, Microsoft’s model, drops one place and ranks fifth. Arena actually assigns the same range of ranks, from 4 to 6, to this model, Nano Banana 2.1, and the new Grok, indicating that the three remain very close at this stage. The new entries push other models down: Reve 2.1 and Reve 2.0 each lose two places, ranking seventh and tenth respectively, as does Muse Image, Meta’s model, now ninth. Seedream 5.0 Pro, ByteDance’s model, falls out of the top 10 to twelfth place.
In this ranking, GPT Image 2.5 Sunburst shows an Elo score of 1,425, GPT Image 2.5 Flare 1,398, and GPT Image 2 medium 1,383. Grok Imagine 2.0 reaches 1,334, MAI-Image-2.6 1,332, and Nano Banana 2.1 1,328. Reve 2.1 obtains 1,302, the August 1 version of Grok Imagine 2.0 1,294, Muse Image 1,274, and Reve 2.0 1,269.
Image Editing: An OpenAI Trio Still Leading
The editing ranking, which measures a model’s ability to modify an existing visual according to instructions, features the same OpenAI trio, with an Elo score of 1,514 for GPT Image 2.5 Sunburst. The new Grok Imagine 2.0 and Nano Banana 2.1 complete the top 5, ahead of MAI-Image-2.6. Krea debuts with Krea 3 Ultra, the most advanced version of its new in-house model, ranking eighth with just over 2,600 votes.
Arena also maintains a separate chart for editing from multiple images, where the instruction is based on several visuals provided as input. The three OpenAI models keep their ranks there, but Krea 3 Ultra ranks fourth, ahead of Nano Banana 2.1, while Seedream 5.0 Pro rises to sixth place. Google fields four models among the top ten, whereas neither SpaceXAI nor Microsoft appear there.
How Arena Establishes Its Rankings
Arena operates by blind duels. For the same instruction, the user receives two images produced by anonymous models and selects the one they prefer, with the names revealed only after voting. These preferences feed into a score modeled on the Elo system, used in chess, where beating a higher-ranked opponent advances one’s score more than a win against a lower-ranked model.
Each score comes with a margin of error and a range of ranks, which narrow as votes accumulate. Recently arrived models, whose position may still change significantly, are marked “preliminary.”
