OpenAI releases ChatGPT Images 2.5 with faster generation and tighter editing controls

OpenAI has released ChatGPT Images 2.5, a broad update to its image generation system that brings a new model to ChatGPT and two new models to the Images API. The company announced the release on September 8, 2026, positioning it as an upgrade for everyday creators, professional teams and developers who need faster image production without giving up control over edits.

The update arrives at a moment when image generation has moved from novelty to routine creative infrastructure. OpenAI says people now create more than 3 billion images each week across ChatGPT Images and GPT-Image models in the API. That scale explains the shape of the release: the new model is not only meant to make prettier images, but to make iterative creative work less fragile. OpenAI says Images 2.5 produces sharper details, more natural lighting and richer textures, while preserving subjects from reference photos more reliably.

One of the main product changes is better precision editing. In earlier image systems, a user could ask for a small change and receive an output that also altered the face, pose, background, typography or composition. OpenAI says the new version is better at changing the requested element while keeping the rest of the image stable. That matters for creators polishing a social image, but it matters even more for production teams that need to keep brand assets, product shots or campaign compositions consistent across multiple rounds of feedback.

The company is also emphasizing multi-turn consistency. Images 2.5 is designed to keep earlier edits intact as a conversation continues, reducing the quality drift that can appear after several rounds of refinement. In practical terms, that makes ChatGPT more useful as an image editing workspace instead of a one-shot generator. A marketer can explore variations, a product team can tune a mockup, and a creator can move from rough concept to finished visual while keeping the original subject recognizable.

OpenAI added new controls inside ChatGPT as part of the same release. Sketch lets users draw a rough reference directly in ChatGPT and pair it with a text prompt. Templates give users a starting structure for common formats such as posters, merchandise and product photos. Users can also place comments on images for focused edits and share prompts so another person can recreate an idea with their own materials.

For developers, the release introduces GPT-Image-2.5 Flare and GPT-Image-2.5 Sunburst. Flare is presented as the default API model for most use cases, with higher quality than GPT-Image-2 and up to 50 percent lower latency. Sunburst is aimed at premium workflows that need tighter control and can tolerate longer generation times. OpenAI says both models are available through the API, while ChatGPT Images 2.5 is rolling out to ChatGPT, ChatGPT Work and Codex users on desktop, mobile and web.

The safety posture remains part of the product story. OpenAI says Images 2.5 continues to use prompt and image checks, C2PA metadata and invisible watermarking to help identify generated media. The larger signal is that image generation is becoming a more managed creative layer: faster, more editable and increasingly integrated into the tools where people already draft, review and ship visual work.