OpenAI Quietly Tests New Image Generation Model 'Mona Lisa 1' on Arena
An unannounced image generator called Mona-lisa-1 has appeared on the Arena platform, and the community has confirmed its OpenAI origin: the company’s official verification service detected hidden SynthID watermarks in the generated images — a signal so rarely false-positive that observers are confident OpenAI is testing the successor to GPT Image 2. Early testers report the model follows complex instructions noticeably more accurately and renders in-image text far better than the current model, clearly outperforming it overall, though some complain about noise and texture artifacts — issues that plagued the previous generation. OpenAI has made no official comment.
Related: OpenAI Previews GPT-5.6 Sol, Meta Releases Image Generator and Video Model Preview