OpenAI rolled out ChatGPT Images 2.5 to ChatGPT, ChatGPT Work and Codex users on September 8, 2026, alongside two new image models for developers, according to Unite.AI, which reported the rollout as reaching all tiers on desktop, mobile and web. The two developer models, named GPT-Image-2.5 Flare and GPT-Image-2.5 Sunburst, are companion releases to the consumer update and are exposed in the API as gpt-image-2.5-flare and gpt-image-2.5-sunburst. OpenAI says people create more than 3 billion images each week across ChatGPT Images and the GPT-Image models in the API.

According to Unite.AI, Images 2.5 produces more natural lighting and richer textures, better preserves subjects from users' reference photos, handles more complex layouts including transparent backgrounds, and holds requested visual styles more consistently. The company also says the model changes only the elements a user asks it to edit while leaving the rest of the image intact, that earlier edits are more likely to carry through as each new instruction builds on previous work, and that it cut image-generation latency by up to 50% compared with Images 2.0.

What is new

  • Typing "@Sketch" in chat opens a canvas for drawing a visual reference, with templates including Poster and Merch
  • Comments can be placed directly on a generated image to request targeted edits instead of rewriting the whole prompt
  • Prompt sharing lets other people rerun an idea with their own photos
  • Developers can call gpt-image-2.5-flare for fast, high-quality everyday generation, or gpt-image-2.5-sunburst for workflows where editing precision matters most

Developer Simon Willison, who maintains the openai_image.py command-line tool for OpenAI's image API, upgraded it to pass reference images through an -i flag and used the new model to add a raccoon scientist to a line chart of agent usage trends, he wrote. Lucky Liao of Manus's evaluation team said Flare delivered high-quality images at two to four times the speed of GPT-Image-2, and the GPT-Image-2.5 models are also available in Adobe Firefly.

On safety, the model's system card, as summarised by Unite.AI, put the unsafe-generation rate at 1.09% for Sunburst and 1.41% for Flare, against 1.64% for ChatGPT Images 2.0, while cautioning that the adversarial test set is not representative of production traffic and that automated policy labels may contain errors. Generated images carry Google DeepMind's SynthID invisible watermarking and continue to include C2PA metadata through the C2PA Conformance Program.

Both API models are priced identically: $8 per million image-input tokens ($2 per million cached), $30 per million image-output tokens, and $5 per million text-input tokens ($1.25 per million cached). Neither report specified separate pricing for the consumer ChatGPT feature, and neither mentioned EU-specific availability terms or euro pricing.