A mysterious new image generation model is beating out those from Midjourney, Black Forest Labs, and OpenAI on the crowdsourced Artificial Analysis benchmark.

Artificial Analysis ranks this model, referred to as "red_panda," as being some 40 Elo points better ranked than the next best of the tested models: Flux1.1 Pro of Black Forest Labs, as ranked by its leaderboard text-to-image results. It uses Elo to measure comparison rank between its many tests using competing models, an application where Elo originated for use by chess players in order to calculate a player's skill compared with others.

The company ranks models for the crowd-sourced Artificial Analysis just like the community AI benchmark Chatbot Arena. When it comes to image models, Artificial Analysis picks up two models and feeds one a unique prompt. It then shows both the prompt and the image it received and asks which the viewer feels more accurately represents the prompt.

Granted, there is some bias in this voting process. Because most voters in Artificial Analysis are enthusiasts of AI, their choices may not represent the entire community of users in generative AI.

But at the same time, red_panda is one of the better-performing models on the leaderboard with respect to generation speed, requiring a median of around 7 seconds to generate an image-over 100 times faster than OpenAI's DALL-E 3.

So, where is the red_panda? Which company is going to make it? And when can we expect to get it? All good questions. AI labs increasingly use community benchmarks to build anticipation ahead of an official announcement, so it mightn't be long before we know.