Overview
- Two OpenAI image models labeled Chestnut and Hazelnut appeared on Design Arena and LM Arena for public blind testing, with early results generally rating them below Google’s Nano Banana Pro on realism, especially human faces.
- Tester demos highlight strengths such as embedding executable code within images and rendering handwritten-style whiteboard text, with several observers suggesting the models build on 4o and improve noticeably over GPT‑Image‑1.
- Analysts note that such arena testing typically precedes major OpenAI launches by one to three weeks, and a Notion entry codenamed “olive‑oil‑cake” plus insider chatter point to a GPT‑5.2 debut as soon as this week.
- Reports say CEO Sam Altman declared a company “code red,” pausing side projects including Sora for roughly eight weeks to prioritize ChatGPT improvements, despite some internal calls to delay the release for more polish.
- Google is preparing follow‑on models described as Nano Banana Flash and Gemini 3 Flash, as Nano Banana Pro continues to draw notice for photorealistic outputs and quick prompt‑based image edits shown in recent demos.