OpenAI's three Alpha models, codenamed Aurora, made a fleeting appearance on the API before being swiftly taken down. Previously, the tech community was under the widespread impression that the Aurora series were premium code models, geared towards high-speed reasoning. Yet, the truth is that these three models constitute the bedrock for OpenAI's upcoming image generation technology and remain accessible in the Arena gray-box testing environment, which is utilized for gathering RLHF (Reinforcement Learning from Human Feedback) data.

User tests have unveiled that the Red Aurora shines in crafting intricately composed, brilliantly lit, and polished close-up images. The Green Aurora, on the other hand, leans towards producing natural and laid-back wide-area lifestyle snapshots, boasting a significant lead in win rates during blind testing matches. Meanwhile, the Blue Aurora is adept at generating images teeming with details, capable of churning out creative content and striking comic visuals. Named after the RGB primary colors, the Aurora series models are poised to lay the visual groundwork for Image-3 or the next-gen multimodal engine.

Recently, OpenAI has rolled out a slew of high-profile offerings. In the face of competitors' cutting-edge large model deployments, it still has a pressing need to showcase more groundbreaking accomplishments down the line.