GPT Image 1.5 dethrones Gemini and becomes the best AI image generator

This is demonstrated by the rankings of LMArena, a platform that evaluates more than 250 LLMs from around the world in 8 categories based on user opinions.
December 17, 2025

OpenAI is doing everything possible to avoid being overshadowed by Google. Since the company led by Sundar Pichai presented Gemini 3 Pro and dethroned all leading LLMs from the top positions of the LMArena rankings, Sam Altman activated the “red code”, demanding rapid and game-changing innovation from his team. This led to the release of the ChatGPT version 5.2, and now, the launch of an improved ChatGPT Images model, which promises to deliver precise edits while preserving details and generating images up to four times faster.

Users have enthusiastically welcomed this upgrade, placing GPT Image 1.5 among the leading positions in the “Text-to-Image” and “Image Edit” categories on LMArena, surpassing Gemini 3 Pro Image, which had previously dominated those categories. This is clearly illustrated in the following image:

As Fidji Simo, Chief Executive Officer of OpenAI’s applications, explains, “our new model is faster and better at following detailed instructions, enabling you to perform more precise edits and creative transformations. It maintains the consistency of key elements such as lighting, composition, and likeness between inputs and outputs, so that the results come much closer to what you envisioned.”

Below, we analyze all the improvements of GPT Image 1.5

Careful edits that respect the original photo

The central promise of GPT Image 1.5 is both simple to understand and highly ambitious: modify only what the user requests and leave everything else untouched. The model is capable of maintaining consistency in elements like lighting, composition, and even people’s features, across several rounds of edits on the same image.

This translates into very concrete use cases:

  • Adding or removing elements
  • Combining multiple images
  • Blending styles
  • Rearranging subjects
  • Transposing entire scenes

All of this occurs without sacrificing visual consistency or narrative continuity from one edit to the next. In practice, this enables smooth workflow chains: make one change, then another, and another… and the image continues to “make sense.”

Creative transformations without prompt fatigue

The other major strength of the model is its so-called “creative transformations”: profound changes that add text, alter the layout, or nearly reinvent a scene entirely while still retaining key details. This leap is not only technical but also improves user experience: the new ChatGPT image space includes predefined styles and concepts, allowing users to generate results without composing lengthy prompts.

This significantly lowers the barrier to entry for less experienced users, as they need only select a style (movie poster, fashion advertisement, retro poster, etc.) and let the model do the heavy lifting.

Improved Compliance with Complex Instructions

One of the longstanding complaints about image generators was their selective obedience: they would follow part of a prompt but ignore the rest. GPT Image 1.5 aims to close this gap. OpenAI assures that the model now follows instructions “more reliably and in greater detail” compared to the previous version, which is evident in examples featuring lengthy and structured specifications (such as 6×6 grids where each cell contains a specific object).

This advancement opens the door for complex compositions where relationships between elements are critical, from infographics to scenes filled with objects positioned with precision. For designers and product teams, it means that the generated image aligns more closely with the original briefing and less with the arbitrary interpretation of the machine.

A major leap in text rendering

Text within images. GPT Image 1.5 represents a further advancement in the clarity and legibility of dense or small text, which is crucial for pieces such as posters, simulated interfaces, or informational graphics. In the provided examples, the model can reproduce complex markdown (headlines, subtitles, tables with numbers and percentages) as if it were a genuine newspaper mock-up.

Furthermore, OpenAI demonstrates how it can reformat the same content to headline other releases, such as its own announcement for GPT‑Image‑1.5.

A new creative space within ChatGPT

Beyond the model itself, ChatGPT now features a dedicated space exclusively for images, accessible from the sidebar on both the web and mobile versions. From there, users may explore:

  • Predefined filters
  • Creative suggestions that update based on current trends
  • The ability to upload an image once and reuse it multiple times
  • Simultaneous generation of multiple images without disrupting the workflow

Quality enhancements: more faces, less “weirdness”

OpenAI is also showcasing improved management of numerous small faces within the same scene, producing images that are more “natural” and displaying fewer visual artifacts in challenging scenarios such as reflections or glare. In comparative testing, complex scenes (for example, a crowded London street during the 1970s) appear sharper, with more clearly defined characters and an overall greater coherence.

In additional examples, the company demonstrates how the model now prevents premature cropping in illustrated posters and achieves outcomes that are more useful for scientific or educational purposes, although it acknowledges that there are still inaccuracies and room for improvement. That is, while this represents a significant advancement, it should not be considered the definitive version (indeed, we should expect further updates and refinements in the future).

GPT Image 1.5 for Developers and Brands

The GPT Image 1.5 version available via API incorporates the same improvements as the new ChatGPT Images, with an emphasis on preserving brands and key visual elements. This makes it particularly attractive for marketing, eCommerce, and creative tools needing to generate complete catalogs—such as variants, scenes, and angles—from a single product photo.

There is also a direct incentive: the cost per image input and output is now 20% lower than with GPT Image 1.0, making it possible to iterate more frequently without straining the budget. OpenAI cites companies such as Wix, Canva, Figma, Envato, and Higgsfield as early adopters, and one testimonial asserts that the model offers high-fidelity images, strong adherence to prompts, and preservation of fine details, placing it “among the benchmark models” in the sector.

Who can use GPT Image 1.5 and how to access it

GPT Image 1.5 is currently being rolled out to all ChatGPT users worldwide, both on the web and the mobile app, without the need to enable any setting or switch models.

It is also available for developers through the API under the name GPT Image 1.5. Access for Business and Enterprise accounts will be implemented progressively.

The previous version of ChatGPT Images will remain available as a custom GPT, but all indications are that this new model will become the new standard moving forward.

Photo: OpenAI

Other articles related to

Published by

Content Manager in Marketing4eCommerce

Stay up to date!

Únete a nuestro canal de Telegram

All you need to know!

Sign up for our newsletter and receive our best articles on eCommerce and digital marketing in your email for free.