OpenAI tests image model amid leadership churn

OpenAI has been testing a new image generator inside ChatGPT and on public leaderboards under code names, even as senior leaders leave and teams are reorganized. (theinformation.com) (storyboard18.com) The Information reported on April 20 that recent images circulating on X and Reddit appear to come from an upcoming OpenAI model, and that “sharp-eyed” users traced tests to some ChatGPT accounts and leaderboard entries with aliases rather than product names. (theinformation.com) OpenAI’s current shipping image product is GPT Image 1.5, which the company said in December 2025 was rolling out to ChatGPT users and the application programming interface, with faster generation and improved editing and text rendering. (openai.com) (community.openai.com) Public benchmarking sites have become a de facto test track for image models because they collect blind human votes on outputs from competing systems. Artificial Analysis’s text-to-image leaderboard currently ranks OpenAI GPT Image 1.5 near the top against Google, Black Forest Labs, ByteDance and xAI models. (artificialanalysis.ai) The product tests are landing during a management reshuffle. Storyboard18 reported on April 18 and April 20 that Srinivas Narayanan, OpenAI’s chief technology officer for business-to-business applications since September 2025, is stepping down next week, while Bill Peebles and Kevin Weil are also departing as the company decentralizes its science division. (storyboard18.com 1) (storyboard18.com 2) Storyboard18 said Narayanan cited personal reasons and a return to India to spend time with his aging parents, and separately grouped his exit with earlier departures including Mira Murati and Shengjia Zhao in a broader leadership shake-up. (storyboard18.com 1) (storyboard18.com 2) For customers, image generation is no longer a toy feature bolted onto a chatbot. OpenAI said in April 2025 that more than 130 million users created over 700 million images in the first week after ChatGPT image generation launched, and it positioned the same capability as an application programming interface for product and design workflows. (openai.com) That makes continuity a procurement issue as much as a research issue. The Information’s April 13 OpenAI org chart said it reflects “the leaders remaining after the departures of senior technical and research leaders in recent years,” underscoring how much institutional knowledge has moved around the company. (theinformation.com) The practical response for software buyers is architectural, not rhetorical: keep image generation behind an internal layer so a product can swap OpenAI for Google or another provider without rewriting the whole application. That approach has become easier as leaderboard-tested models converge on similar features such as prompt following, editing and text rendering. (artificialanalysis.ai) (openai.com) OpenAI is still shipping image products and expanding its business lineup, but the overlap between quiet model trials and visible executive turnover leaves customers watching two things at once: output quality and organizational stability. (openai.com) (theinformation.com)