OpenAI's mystery image models spotted in the wild
Mysterious new image models codenamed Chestnut and Hazelnut just appeared on LM Arena and Design Arena, with users speculating they belong to OpenAI. Testers report sharp world knowledge, realistic selfies, and clean in-image code rendering. Why it matters: If the rumor mill is right, OpenAI may be gearing up to launch its next-generation image system alongside GPT 5.2.
Mystery models are once again sneaking into public benchmark arenas before their official debut.
Two new image models, codenamed Chestnut and Hazelnut, were spotted testing on Design Arena and LM Arena. Testers suspect they could be OpenAI's next-generation image models, potentially slated to drop alongside GPT 5.2 as early as this week or next.
Selfies and clean code rendering
Early testers noted a few standout capabilities from the outputs.
The models show a level of world knowledge comparable to Nano Banana Pro, along with high-quality celebrity selfie generation. More interestingly, testers reported that the models can render legible code directly inside generated images, tackling a long-standing weakness in image generation.
Why it matters
We think this suggests OpenAI is preparing to seriously upgrade its visual generation capabilities to keep pace with rivals.
Precise text and code rendering inside images has been a persistent pain point for creators building UI mockups, infographics, and technical diagrams. It looks like OpenAI wants native image generation tightly integrated into its next model cycle, though since the models are still unconfirmed test candidates, we will have to wait for an official announcement to see if they live up to the hype.
Sources (1)
- x.com(opens in a new tab)Originalx.com