GPT-IMAGE-2 likely testing on LMarena

OpenAI is leveraging the LMarena feedback loop to refine its DALL-E successor before an official release, signaling a major leap in multimodal capabilities. The anonymized "tape" series models significantly outperform competitors like Google's Nano Banana in complex prompt adherence and text generation. Significant gains in anatomical detail and skin texture suggest a more robust diffusion process or a move toward higher-quality training datasets. Direct model attribution as "OpenAI" within the chat interface confirms the source despite the hidden labels during blind testing. The move back to LMarena suggests OpenAI is prioritizing leaderboard performance to recapture the state-of-the-art title in the image-gen category.