OpenAI's ChatGPT Images 2.0 tested with 'Where's the raccoon' challenge, outperforming competitors

OpenAI launched ChatGPT Images 2.0 on 21st April 2026, with CEO Sam Altman claiming the improvement from the previous version equals the leap from GPT-3 to GPT-5. A test was conducted using a 'Where's Waldo' style prompt requesting an image of a raccoon holding a ham radio. The older gpt-image-1 model failed to produce a recognizable raccoon, while Claude Opus 4.7 incorrectly identified one due to an instruction card. Google's Nano Banana 2 via Gemini generated an obvious raccoon in an 'Amateur Radio Club' booth with a 'W6HAM' callsign pun, but Nano Banana Pro delivered poor results. Using an updated script with the new model at maximum settings (3840x2160 resolution, high outputQuality) produced a 17MB PNG with a clearly visible raccoon in the bottom left, costing approximately 40 cents for 13,342 output tokens. The model now surpasses Gemini in this specific test. A Hacker News user named rizaco demonstrated that these models cannot reliably solve their own puzzles by asking ChatGPT to circle the raccoon in a previously failed image.