OpenAI Releases ChatGPT Images 2.5: Major Upgrades to Image Editing and Multi-Turn Dialogue

OpenAI has launched ChatGPT Images 2.5, the latest version of its image generation model. According to official data, OpenAI's image generation models have collectively produced over 3 billion images across ChatGPT Images and the API. This update delivers significant improvements in multi-turn instruction following, response speed, and reference photo subject preservation.

Sunburst and Flare: A Dual-Model Differentiation Strategy

This update introduces two new model IDs: gpt-image-2.5-sunburst and gpt-image-2.5-flare. According to OpenAI's official documentation, the two models serve distinct use cases:

• Sunburst: Designed for high-precision editing workflows, ideal for professional scenarios requiring fine-grained control over image details
• Flare: Focused on fast response times and high-quality output, suited for everyday image generation needs

This differentiated positioning allows users to flexibly choose between processing speed and editing precision based on their needs. For professional work requiring repeated iterative adjustments, Sunburst is the better fit.

Reference Image Feature: Preserving the Original While Adding New Elements

ChatGPT Images 2.5 marks a major breakthrough in reference photo subject preservation. Developer Simon Willison demonstrated this capability through a real-world example: he upgraded his openai_image.py CLI tool and used the following command to add a cartoon character to an existing data chart:

uv run https://tools.simonwillison.net/python/openai_image.py \\
  'add a raccoon scientist studying the chart thoughtfully' \\
  -i https://static.simonwillison.net/static/2026/openai-agent-usage.webp \\
  -m gpt-image-2.5-sunburst

The generated image fully preserved all data points and annotations from the original line chart, while naturally incorporating a raccoon scientist wearing glasses and a lab coat — complete with environmental details like an OpenAI logo mug, a printed chart, and AI-related books. This demonstrates the model's ability to execute complex editing instructions while maintaining the integrity of the original content.

Dual Optimization: Multi-Turn Dialogue and Response Speed

This update places particular emphasis on improving instruction-following ability across multiple turns. Users can refine and adjust image generation requests multiple times within the same session, with the model better understanding context and maintaining consistency. This is especially valuable for creative workflows that require repeated iteration.

The improvement in response speed is equally noteworthy. Faster generation times enhance the overall user experience — particularly when quickly validating creative concepts or running batch image generation tasks, where the efficiency gains will be even more pronounced.

Broader Impact on the AI Image Generation Industry

The 3 billion image milestone underscores OpenAI's leading position in the image generation space. With this update's dual-model strategy, OpenAI demonstrates a growing maturity in addressing differentiated user needs within a commercial AI product.

From a technical perspective, the enhanced reference image preservation capability addresses one of AI image generation's core challenges: how to maintain the integrity and accuracy of existing content while adding new elements. This breakthrough will expand the applicability of AI image generation in professional domains.

For developers, the improved API usability and CLI tool extensibility lower the barrier to technical integration, making image generation capabilities easier to embed into various automated workflows. This trend toward tooling will further accelerate the deep adoption of AI image generation across vertical industries.