June 16, 2026, 14:30

OpenAI takes top three spots when users pick AI images blind

AI-generated from the sources listed, with automated source checks.

On the Arena leaderboard for AI image models, OpenAI leads by a wide margin over Google and Grok. Claude has no image model among the top ten.

Which AI makes the best-looking images? On the big Arena leaderboard (formerly LMArena), the answer right now is OpenAI. The leaderboard was updated on October 26, 2026.

Arena is based on blind voting. Users are shown two images to choose from, without knowing which model made them. In total, the leaderboard is based on 6,872,557 votes.

OpenAI takes the top three spots

No. 1 is OpenAI's GPT-Image-2.5 Sunburst with a score of 1427 (±5) and 40,323 votes. No. 2 is GPT-Image-2.5 Flare with 1401, and No. 3 is GPT-Image-2 (Medium) with 1381.

In fourth place is Microsoft's MAI-Image-2.6 with 1329. Google's Gemini Nano Banana 2.1 is fifth with 1325, but that score is based on only 5,290 votes and is therefore less certain (±9). The gap between first place and Google's best model is about 100 points.

Grok's Imagine Image 2.0, developed by SpaceXAI, is sixth with 1317. A version from August 1, 2026 is eighth with 1288. Reve 2.1 is seventh, and Meta's Muse Image is ninth.

There is no image model from Anthropic, the company behind Claude, among the top ten.

A single test with a cricket banner

A developer on DEV Community tested ChatGPT, Claude, Gemini and Grok on the same task: graphics for a cricket club, both a static banner and a short animated video. It is one person's test with one task, and it is not a blind test.

ChatGPT delivered two finished, professional banners. Claude produced HTML code instead of an image. Grok delivered a messy banner, while Gemini declined the task, citing safety rules.

In the video portion, Grok did best with a six-second clip, but the video had the wrong phone number. Gemini's video had incoherent text.

The author of the test concludes that the best AI for code is not automatically the best for graphics. His advice is to choose the tool by task: ChatGPT for static images and Grok for animation.

What does it mean for you?

The leaderboard measures image models, not the chatbots as a whole, and it changes continuously. Placements close to each other are uncertain. However, the large gap between OpenAI and the rest suggests that OpenAI currently has a clear lead in users' blind choices.