Updated
Updated · kucoin.com · Sep 9
OpenAI Launches GPT-Images-2.5, Cutting Latency by 50% and Splitting API Into 2 Models
Updated
Updated · kucoin.com · Sep 9

OpenAI Launches GPT-Images-2.5, Cutting Latency by 50% and Splitting API Into 2 Models

3 articles · Updated · kucoin.com · Sep 9

Summary

  • OpenAI said GPT-Images-2.5 is now live for all ChatGPT, ChatGPT Work and Codex users — including the free tier — with image generation latency reduced by up to 50%.
  • The upgrade focuses on tighter editing control: users can change only selected regions, preserve composition and subjects across repeated edits, and mark areas directly on images for revision.
  • Four product additions broaden input beyond text, including Sketch for hand-drawn prompts, templates for posters and product images, and prompt sharing for reusable styles.
  • For developers, OpenAI split its image API for the first time into Flare for faster, high-volume generation and Sunburst for slower, higher-precision creative work; Adobe said it is integrating the model into Firefly.
  • OpenAI said ChatGPT Images and the GPT-Image API now process 3 billion images a week, underscoring how central image generation has become to its broader multimodal push.

Insights

Why are some users calling OpenAI's massive new image upgrade flat and generic despite promises of hyper-realistic detail?
If AI metadata can be easily stripped, how will we ever truly distinguish human art from a flawless Sunburst generation?