← Back to feed

OpenAI now processes more than 3 billion images per week across ChatGPT Images and the GPT-Image API. The volume justified a full model generation upgrade. ChatGPT Images 2.5, released September 8, 2026, ships with two distinct API models, a 50% latency improvement over Images 2.0, and a set of new platform features aimed at tightening the gap between what users describe and what the model produces.

Two API models, distinct roles

The API now offers:

GPT-Image-2.5 Flare — the standard choice. Delivers the full Images 2.5 quality and editing improvements with the speed benefits. Intended for production workflows, iterative editing, and integration into products where response time matters.

GPT-Image-2.5 Sunburst — the precision tier. Trades generation speed for extra detail fidelity. Suited for work where visual accuracy is more important than throughput: print-quality assets, detailed product photography, controlled creative production.

The naming mirrors OpenAI’s recent approach of offering a fast default and a premium-precision alternative within the same model generation. Sunburst and Flare entered the Arena Text-to-Image, Image Edit, and Multi Image Edit leaderboards on September 8.

What changed in the model

The Images 2.5 generation delivers four substantive improvements over Images 2.0:

Reference photo fidelity: The model is better at preserving subject identity across transformations. Distinctive features — face structure, object geometry, brand elements — carry through more reliably when the source image changes style, background, or composition.

Precision editing: The model is better at editing only what was asked for, leaving the rest of the frame unchanged. Earlier versions tended to drift across the full image when given a local edit instruction.

Multi-turn consistency: In extended ChatGPT conversations with multiple edit rounds, Images 2.5 maintains prior edit state more reliably. Earlier changes are less likely to degrade as subsequent instructions accumulate.

Generation speed: Up to 50% lower latency versus Images 2.0. The improvement applies to both initial generation and subsequent edit rounds.

New ChatGPT platform features

Alongside the model, OpenAI shipped three new creation features in ChatGPT:

• Sketch: Draw directly in ChatGPT as a reference image for the final output. Replaces the approximation of describing a spatial layout in text.
• Templates: Starting points for common formats including flyers and product photos.
• Image comments: Annotate specific regions of an image to direct edits with more precision than a full-sentence prompt.

A prompt-sharing feature lets users share the exact prompt used to generate an image, enabling others to apply the same prompt to their own reference photos.

Availability

Images 2.5 is available to all ChatGPT, ChatGPT Work, and Codex subscribers across desktop, mobile, and web as of September 8. The API models GPT-Image-2.5 Flare and GPT-Image-2.5 Sunburst are available to developers immediately.

The Arena leaderboard additions on the same date will generate comparative data against existing top-ranked image models — currently GPT-Image-2 and xAI Imagine Image 2.0 dominate image editing, with Microsoft MAI-Image-2.6 at number two. First Arena scores for both Images 2.5 variants should appear within days.