OpenAI released GPT Image 2.5 on September 8, 2026, rolling it out across ChatGPT, ChatGPT Work, and Codex on desktop, mobile, and web, alongside two new models in the API. It's the successor to GPT Image 2.0, built around faster generation, more precise editing, and better handling of reference photos.

What Is GPT Image 2.5?

GPT Image 2.5 is OpenAI's latest AI image generation and editing model, available both inside ChatGPT and through the API as two distinct variants: Flare and Sunburst. It generates images from text prompts, edits existing images with targeted changes, and works from reference photos to place familiar subjects into new settings, styles, or compositions.

Key Features of ChatGPT Images 2.5

• Better Instruction Following and Style Consistency: GPT Image 2.5 handles more complex visual instructions without losing the thread partway through. Images built around real-world information come out more accurate, and layouts with several elements, including ones combined with transparent backgrounds, hold together more coherently than in GPT Image 2.
• Precise Multi-Turn Editing: Instead of regenerating an entire image and hoping the parts that worked survive, GPT Image 2.5 changes only the region specified in the edit instruction, leaving the rest of the composition untouched. Multi-turn editing holds up better across several rounds too, so a sequence of small revisions doesn't slowly drift away from the original image.
• Reference Image Fidelity: GPT Image 2.5 accepts up to 16 reference images in a single edit, and it holds onto what makes a subject recognizable far more consistently than GPT Image 2 did. This matters directly for product photography and brand work, where the point is that the viewer recognizes exactly what they're looking at, not just something similar to it.
• Improved Text and Label Rendering: Short labels, headlines, and infographic text come out noticeably cleaner than in GPT Image 2, which matters for anything combining images with on-image copy: product mockups, infographics, and marketing visuals with a headline baked into the frame.
• New Creation Tools Alongside the Model: ChatGPT's image interface gained a Sketch feature for rough visual input, Templates for repeatable formats, the ability to comment directly on a generated image, and shareable prompts, so a specific generation setup can be handed to someone else instead of described from scratch.
• Five Quality Tiers, Including Two New Ones: Quality now runs across five settings: low, medium, high, xhigh, and max, plus an automatic option that picks a sensible default. Xhigh and max are new to this generation.

Try ChatGPT Images 2.5 on ImagineArt

Flare vs. Sunburst: The Two Models

GPT Image 2.5 ships as two API models. Both models bill at the same token rate through the API, so the choice between them comes down to whether a project needs speed or precision, not cost.

• GPT Images 2.5 Flare is the default. It delivers higher-quality output than GPT Image 2 while running at up to 50% lower latency, making it the choice for creator and social content, product listings, visual search, rapid prototyping, and any workflow generating images in volume.
• GPT Images 2.5 Sunburst trades speed for precision. It spends more time per image, which shows up in tighter control over edits and stronger detail preservation, the tier suited to campaign visuals, product photography, architectural imagery, or a hero frame where every detail needs to hold up.

Technical Specs

• Maximum resolution: 3840 x 2160, with a maximum edge of 3840 pixels
• Standard sizes: 1024x1024, 1536x1024, 1024x1536, plus custom sizes in multiples of 16 pixels
• Aspect ratios: 1:3 to 3:1
• Quality tiers: low, medium, high, xhigh, max, or automatic
• File formats: PNG, JPEG, and WebP, with support for transparent backgrounds
• Images per request: up to 10
• Reference images per edit: up to 16
• Provenance: C2PA metadata plus an invisible SynthID watermark
• Not supported: fine-tuning, function calling, or structured outputs

Use Cases for ChatGPT Images 2.5

• Infographic creation and iterative refinement.
• Product photography and e-commerce imagery.
• High-volume content pipelines.
• Brand and marketing visuals with on-image text.
• Campaign visuals and hero images.
• Transparent assets for layered design work.
• Rapid prototyping and layout exploration.

How to Use GPT Image 2.5 on ImagineArt

GPT Image 2.5 is also accessible through ImagineArt AI Image Generator, alongside other image models, without setting up a separate API account or managing token-based billing directly.

• Go to the ImagineArt dashboard and open the AI Image Generator from the sidebar.
• From the model dropdown, select GPT Image 2.5.
• Enter a text prompt to generate a new image, or upload a reference image to edit an existing one.
• Set the aspect ratio, resolution, and quality tier needed for the output.
• Generate the image, then download it or continue editing it in the same dashboard.

Try ChatGPT Images 2.5 on ImagineArt

Pricing and Access

GPT Image 2.5 is available directly inside ChatGPT for existing subscribers. For standard processing:

• Image input: $8.00 per 1M tokens
• Image output: $30.00 per 1M tokens
• Text input: $5.00 per 1M tokens
• Text cached input: $1.25 per 1M tokens

On ImagineArt, ChatGPT Images 2.5 (both Flare and Sunburst) consumes 3 credits/image for low quality and 1K resolution. You can try GPT Images 2.5 for free on ImagineArt using the daily 100 free credits.