At this point, Google and OpenAI are battling it out neck to neck in terms of AI models. Earlier today, we reported that Google is finalizing an affordable image generation model that offers somewhat similar image quality as the Nano Banana Pro model. Here, the image generation model in question is the Nano Banana 2 Flash, and it will reportedly be powered by Gemini 3 Flash.
OpenAI is reportedly testing a new AI image model
Now, OpenAI appears to be testing new AI image models codenamed Chestnut and Hazelnut. The model has been spotted in Design Arena and LM Arena over the past 24 hours. And, if the report by @marmaduke091 on X is accurate, OpenAI’s next major imaging upgrade may debut alongside the GPT-5.2 release, which is expected today.
According to early testers, the reported OpenAI model shows notable improvements in world knowledge, as much so that it has been compared to Google’s Nano Banana Pro. Early samples also show that the model generates celebrity-style selfies with accuracy and photorealism close to that competitor. However, it’s still not on the par with the Nano Banana Pro model. I’ve attached some of the samples below, take a look:
That’s not all; the alleged OpenAI model can render code inside images more reliably, something that has historically challenged image models. This suggests OpenAI might be optimizing architecture for technical illustrations, UI mock-ups, and developer-focused content. As usual skepticism remains among users, as some users noted that the faces in sample images appear “plasticky.”
The tested model allegedly uses GPT-4o
The alleged OpenAI image model that’s being tested right now is rumored to be based on an earlier GPT-4o-based imaging system. However, the new image model is still a significant jump over GPT-Image-1, which launched earlier this year.
“I think it’s still based on 4o,” one tester wrote on X. “However, it will be a huge jump from GPT-Image-1… still not quite on the level of Nano Banana Pro, but we need to wait for the full release.” Now the model has been spotted in the wild, with references appearing across testing platforms, an official announcement appears increasingly likely.
