OpenAI testing new "Aurora" GPT-Image model on Arena
OpenAI is quietly testing what may be its next image generation model, currently referred to as "Aurora." Three alpha variants of the model surfaced briefly in OpenAI's API before being removed, a pattern consistent with how the company has historically staged early-access rollouts and internal capability checks.
The models have since been spotted in Arena-style evaluations - blind side-by-side comparisons where users rate outputs without knowing which model produced them. This format has become a common method for AI labs to gather preference data at scale, and OpenAI has used similar approaches before to inform decisions around model promotion and deployment timing.
Aurora's relationship to OpenAI's existing image lineup is not yet fully clear. The company currently offers GPT-Image-1, which powers the image generation features in ChatGPT and the API. The "Aurora" naming and the presence of multiple alpha variants suggest iterative testing rather than a single candidate model, which may indicate OpenAI is exploring different capability or efficiency trade-offs before settling on a final configuration.
No official announcement has been made, and OpenAI has not confirmed the Aurora name or provided details about the model's architecture or intended release timeline. Given the early-stage nature of the API appearances, a public release is likely still some distance away - but the Arena testing phase typically signals that a model is far enough along for structured external feedback.
