OpenAI is testing three new image generation models codenamed Aurora in its Arena platform. Aurora-red-alpha and aurora-green-alpha show differences in speed and quality, while aurora-blue-alpha remains untested.
TL;DR
- OpenAI is testing three new image generation models: aurora-red-alpha, aurora-green-alpha, and aurora-blue-alpha.
- Aurora-red-alpha is faster but lower in quality compared to aurora-green-alpha, which is slower but produces better images.
- The models may be part of the upcoming GPT-Image-3, but this is not yet confirmed.
What happened
Three new model references, aurora-red-alpha, aurora-green-alpha, and aurora-blue-alpha, briefly appeared in OpenAI's API and were subsequently spotted in Arena tests generating images. The references were pulled shortly after, suggesting an accidental rollout. According to Lumina Bench, all three models appear to be image models currently under testing on Arena. The first two models, aurora-red-alpha and aurora-green-alpha, have been tested, with the former being faster but lower in quality, and the latter being slower but producing better images. The blue model hasn't been tested yet.
Separately, three new entries named Aspen, Cypress, and Maple appeared on Design Arena about an hour after the API references surfaced. However, there is no direct tie between these names and the Aurora models.
Why it matters
This development is significant for AI/ML developers and startup founders as it hints at potential advancements in image generation technology. The testing of multiple models suggests OpenAI's commitment to improving both speed and quality in its image generation capabilities. For tech investors, this could indicate a competitive move in the AI space, potentially influencing market dynamics.
The lack of documentation or pricing for the Aurora models means that the full scope and implications of these tests are still unclear. Additionally, the focus on image generation models comes at a time when users are also anticipating new frontier models from OpenAI, such as the recently launched GPT-6 with Intelligent UI and the upcoming gpt-rosalind-discovery.
Key facts
- Three new model references, aurora-red-alpha, aurora-green-alpha, and aurora-blue-alpha, were spotted in OpenAI's API and Arena tests.
- Aurora-red-alpha is faster but lower in quality compared to aurora-green-alpha, which is slower but produces better images.
- The models may be part of the upcoming GPT-Image-3, but this is speculative.
- Three new entries named Aspen, Cypress, and Maple appeared on Design Arena, but their connection to the Aurora models is unclear.
- OpenAI has previously released multiple image generation models, such as GPT-Image-2.5 Flare and GPT-Image-2.5 Sunburst.
- There is no documentation or pricing available for the Aurora models yet.
Context
OpenAI's testing of multiple models is not unprecedented. The company has previously released multiple image generation models, such as GPT-Image-2.5 Flare for speed and GPT-Image-2.5 Sunburst for precision. Additionally, OpenAI ran three tape-themed Image V2 variants through LM Arena back in April.
The focus on image generation models comes at a time when OpenAI is also working on other advancements, such as the recently launched GPT-6 with Intelligent UI and the upcoming gpt-rosalind-discovery. Meanwhile, competitors like Anthropic are preparing to launch new models, such as Claude Fable 5.5, indicating a busy period for AI developments.