A mysterious latest image generation model beats models from Midjourney, Black Forest Labs and OpenAI within the crowdsourced artificial analytics benchmark.
The model called “red_panda“is around 40 Elo points ahead of the second best model, Flux1.1 Pro from Black Forest Labs, in Artificial Analysis’s text-to-image leaderboard. Artificial Analysis uses Elo, a rating system originally developed to calculate the relative skill levels of chess players, to check the performance of different models tested.
Similar to the community AI benchmark Chatbot Arena, Artificial Analysis evaluates models through crowdsourcing. For image models, artificial evaluation randomly selects two models and provides them a singular prompt. The prompt and resulting images are then displayed, and users select which one they think higher reflects the prompt.
Admittedly, there may be some bias on this voting process. Artificial Analysis voters are largely AI enthusiasts, and their decisions may not reflect the preferences of the broader community of generative AI users.
But red_panda can be probably the most powerful models on the leaderboard when it comes to generation speed. The model takes about 7 seconds on average to generate a picture – over 100 times faster than OpenAI's DALL-E 3.
So where does red_panda come from? Which company made it? And when can we expect the discharge? All good questions. AI laboratories are increasingly using community benchmarks for this purpose Arouse anticipation However, the announcement continues to be imminent, so it may not be long before we discover out.