OpenAI launches new GPT Image 1.5 model optimized for image editing – News

OpenAI Group PBC today launched GPT Image 1.5, a new artificial intelligence model optimized for image generation tasks.

The algorithm is rolling out a few weeks after Google LLC introduced a new image generator of its own. Nano Banana Pro, as the model is called, is better than its predecessors at rendering text and implementing instructions from follow-up prompts. GPT Image 1.5 likewise introduces improvements in those areas along with several other optimizations.

Developers can access GPT Image 1.5 via an application programming interface. The API pricing is 20% lower than for OpenAI’s previous image generation model. Nontechnical users, in turn, can interact with the algorithm via ChatGPT’s image generation feature.

According to OpenAI, ChatGPT can create images up to four times faster than before thanks to the model switch. That suggests GPT Image 1.5 is more hardware-efficient than its predecessor. Reducing a model’s hardware requirements not only boosts processing speeds but also lowers costs: The less time an AI takes to complete a task, the less electricity it consumes.

When AI models edit an image, they sometimes modify details that the user didn’t ask them to change. That’s not always an issue for consumer use cases, but can be a challenge in commercial projects. A brand using ChatGPT to resize a product image wouldn’t want the chatbot to remove its logo.

OpenAI says GPT Image 1.5 is better than its predecessor at retaining important image details while making changes. It’s also more adept at carrying out elaborate, multiple-step edits. For example, a user could ask the model to place objects from three different drawings in a single image and then change the style in which the objects are illustrated.

Text rendering, one of Nano Banana Pro’s major selling points, was also a focus for OpenAI’s researchers while they trained GPT Image 1.5. The ChatGPT developer says that the model is better than its predecessor at rendering small, dense text. That makes it more suitable for tasks such as generating infographics.

OpenAI disclosed in a blog post that GPT Image 1.5 also has certain limitations. According to the company, the model provides limited support for certain drawing styles and sometimes makes mistakes when generating images that require scientific knowledge. However, it fares significantly better in the latter area than its predecessor . 

The model is accessible through a new interface section in ChatGPT. The section includes pre-configured image filters and prompt suggestions that OpenAI plans to periodically refresh.

GPT Image 1.5 is rolling out a few days after the company upgraded ChatGPT to GPT-5.2, its newest foundation model. OpenAI says that the latter model is better at scientific tasks than earlier algorithms. In an internal test, GPT-5.2 set a record on a AI benchmark that comprises college-level math questions.

OpenAI today released an AI benchmark of its own FrontierScience. It’s designed to measure how well foundation models perform scientific tasks. The benchmark comprises more than 700 textual questions spanning physics, chemistry and biology.

Image: OpenAI