OpenAI is Reportedly Working on New AI Image Models Codenamed "Chestnut" & "Hazelnut"

It’s been almost over 7 months since OpenAI debuted its latest AI image model, gpt-image-1. With increased competition with Nano Banana Pro, OpenAI appears to be also testing new AI image models with codenames, Hazelnut and Chestnut. These findings were shared by @marmaduke091 on X. 

The user also suggested that the model has been spotted in Design Arena and LM Arena over the past 24 hours. And, if the report is accurate, OpenAI’s next AI model might debut alongside the GPT-5.2 model, rumored to be launched today. We are yet to see if that happens or not.

As reported, the new image models from OpenAI are great in terms of world knowledge and on par with Google’s Nano Banana Pro. Early samples of the image generated using the alleged OpenAI’s new AI image model depict celebrity-style selfies with accuracy and photorealism close to that of its rival. That being said, it’s still far from beating the accuracy and detail you get from the Nano Banana Pro model. 

One of the interesting claims being made by early testers is that the new image model from OpenAI renders code inside images more reliably. If this is true, it’s a great feat. That being said, many users on X still feel that the celebrities’ faces from the above sample images look “plasticky.” When I first saw it, I felt the same. That isn’t the case with images generated using Google’s Nano Banana Pro. 

The alleged OpenAI image model that’s being tested right now is rumored to be based on an earlier GPT-4o model and not the latest gpt-image-1. All that said, we still need to wait for the official release to tell us something concrete. 

What do you think about the alleged OpenAI image models? Do let us know in the comments below. 

Sample images via X/@marmaduke091