Same-prompt render test puts codenamed Gemini 4 a release cycle ahead of Opus 5.5
Using one prompt to render the same city, the author says Gemini 4 draws scenes the other model can only describe, and ranks Gemini 4 Pro and Opus 5.5 on top with Astra in between.
Same prompt, same city, two models. One of them renders a scene the other one can only describe. If the Gemini 4 numbers are real, Google just pulled off the biggest comeback this industry has seen. The model is sitting in Arena under a codename. Google has confirmed nothing. Everyone testing it blind keeps reporting the same thing: the gap is not incremental. My own read after putting four of them side by side: → Gemini 4 Pro and Claude Opus 5.5 are in a tier of their own right now → 3.8 Flash punches far above what its speed tier should allow → GPT-6 Astra lands in between, ahead of Flash, behind the top two Watch the left frame, then the right one. That difference is one release cycle wide. Attribution is unconfirmed and my ranking is one person's impression. The clip is not.