OpenAI Tests Next-Generation Image Model gpt-image-2, Focused on Complex Charts and Realism

According to monitoring by Dongcha Beating, OpenAI’s upcoming next-generation image generation model has been tested among some ChatGPT users and has appeared on third-party lists under the codename gpt-image-2. The model is expected to be officially released in the coming weeks. Insiders mentioned two key improvements. First, the capability for complex charts has significantly enhanced. For instance, it can accurately place sticky notes or text at specific locations on a whiteboard, a task that the previous model struggled with. Second, the output no longer has an ‘AI feel’; the overly perfect lighting and skin-smoothed faces that were previously recognized flaws have been addressed. The new model is driven by OpenAI’s pressure for user growth. ChatGPT’s weekly active users have recently plateaued at around 920 million, missing the goal of reaching 1 billion by the end of 2025. Sam Altman issued an internal ‘code red’ last year, triggered by the popularity of Google’s image model Nano Banana, with improvements in image generation being prioritized as one of the ‘code red’ initiatives. OpenAI aims to replicate the ‘Ghibli moment’ of early 2025, when millions of users flocked to ChatGPT to transform their and their friends’ photos into Ghibli-style images. With enhanced text rendering capabilities, images can also be used more reliably in advertising and educational charts. This is one direction in which OpenAI seeks to differentiate itself from Anthropic, which has recently surpassed OpenAI in revenue with its programming models.