OpenAI released ChatGPT Images 2.5 on Tuesday, September 8, cutting image generation latency by up to 50% compared with Images 2.0 and introducing a freehand drawing input called Sketch1,3.
Sketch lets users draw directly inside ChatGPT by typing @Sketch into the chat box, which opens a canvas for doodling. ChatGPT then generates an image based on the sketch and accompanying text instructions. The Verge's brief test turned a mouse-drawn cat doodle into a realistic photo. Users can also now leave comments directly on parts of generated images.
Beyond the new input mode, ChatGPT Images 2.5 improves instruction-following across multiple conversation turns and is reportedly better at preserving subjects in reference photos2. Two new model IDs are available in the API: gpt-image-2.5-sunburst and gpt-image-2.5-flare. OpenAI's guidance positions Sunburst for workflows where editing precision matters most, and Flare for fast, high-quality everyday image generation.
OpenAI said its image generation models have been used to create more than 3 billion images across ChatGPT Images and the GPT-Image models in the API.
The API availability prompted immediate tooling updates. Developer Simon Willison upgraded his openai_image.py CLI tool to support passing in one or more reference images alongside text prompts, demonstrating the Sunburst model by compositing a raccoon character onto an existing chart image.
ANALYSIS The Sketch feature turns ChatGPT's image generator into a spatial interface rather than a purely text-driven one, giving users a way to communicate composition and layout that natural language handles poorly. The dual-model API strategy, splitting precision editing from fast generation, gives developers a latency-quality tradeoff to tune per use case rather than accepting a single default.
The release comes one day after OpenAI said it had reached its goal of building an automated research intern capable of handling multi-day research tasks[2].