AI Catchup

OpenAI Launches ChatGPT Images 2.5 and GPT-Image-2.5 API Models

By 7 min read

OpenAI launched ChatGPT Images 2.5 with sharper image fidelity, faster generation, more precise multi-turn edits, Sketch, templates, and comment-based editing. The same release adds GPT-Image-2.5 Flare for fast everyday API generation and Sunburst for more precise, premium creative workflows.

OpenAI launched ChatGPT Images 2.5 on September 8 with sharper details, faster generation, more precise editing, and new controls for creating and sharing images. The same release adds GPT-Image-2.5 Flare and GPT-Image-2.5 Sunburst to the API: Flare is optimized for fast everyday generation, while Sunburst targets detailed creative work. (OpenAI; OpenAI on X; OpenAI Developers on X)

Key Takeaways

  • ChatGPT Images 2.5 is the user-facing release. OpenAI says it improves image fidelity, generation speed, and editing consistency across multiple turns. (OpenAI)
  • Sketch turns a drawing into an image reference. You can type @Sketch in ChatGPT and use a rough drawing to guide the final image. (OpenAI)
  • Templates and comments make editing more directed. Templates help you start with formats such as posters and merch, while comments let you focus an edit on a specific part of an image. (OpenAI)
  • Flare and Sunburst are the API models. Flare is the default choice OpenAI recommends for most applications; Sunburst is for workflows that need more editing precision. (OpenAI; OpenAI image generation guide)
  • Both API models use GPT Image 2 token rates and add two quality settings. Image output is $30 per million tokens on Flare and Sunburst alike; both add xhigh and max above high, default to auto, and pin to the 2026-09-08 snapshot. (GPT Image 2.5 Flare model page; OpenAI image generation guide)
  • Availability spans OpenAI's main user surfaces. Images 2.5 is rolling out to ChatGPT, ChatGPT Work, and Codex across desktop, mobile, and web, while both API models are available in the API. (OpenAI)

What Changed In ChatGPT

ChatGPT Images 2.5 is designed to keep more of the image you want while changing only the part you asked to change. OpenAI says it improves natural lighting, richer textures, recognition of subjects in reference photos, and instruction following across multiple editing turns. (OpenAI)

The release also adds four workflow features:

  • Sketch: Draw directly in ChatGPT and use the drawing as a visual reference. Type @Sketch to start.
  • Templates: Start from formats such as posters or merch rather than a blank prompt.
  • Comments: Place comments on an image to direct a focused edit.
  • Shared prompts: Share the prompt used to create an image so another person can try it with their own photos and details.

OpenAI reports that Images 2.5 reduces generation latency by up to 50% compared with Images 2.0. Treat that as OpenAI's reported comparison, not as a guarantee for every prompt, device, or workload. (OpenAI)

The Two API Models

OpenAI is releasing two GPT-Image-2.5 models for developers. Both carry the quality, editing, and speed improvements from Images 2.5 into API workflows, but they are aimed at different operating points. (OpenAI; OpenAI API changelog)

ModelBest fitTradeoff
gpt-image-2.5-flareFast, high-quality everyday generation, creator and social content, product experiences, visual search, prototyping, and high-volume generationOptimized for speed and general use
gpt-image-2.5-sunburstDetailed editing, campaign creative, polished product imagery, and other premium visual workflowsLonger generation times in exchange for tighter editing precision

OpenAI describes Flare as the default choice for most applications. Its product announcement says Flare delivers higher-quality images than GPT-Image-2 at 50% lower latency, while Sunburst adds more control for detailed edits. (OpenAI)

The API supports both the Images API and image generation in the Responses API. Use the Images API for a single generation or edit, and use the Responses API when image generation belongs inside a conversational or multi-step flow. OpenAI's developer guide documents both paths and the two model names. (OpenAI image generation guide)

Pricing, Quality Settings, and Snapshots

Flare and Sunburst cost the same per token, and both match GPT Image 2. The image generation guide says: "Both GPT Image 2.5 models use the same token rates: $8 per million image input tokens, $2 per million cached image input tokens, $30 per million image output tokens, $5 per million text input tokens, and $1.25 per million cached text input tokens." Text output is not billed, because the models output images. As of October 2026 the Flare and Sunburst model pages add a Batch rate: image output costs $30 per million tokens with Standard processing or $15 with Batch processing. (OpenAI image generation guide; OpenAI API pricing)

SettingGPT-Image-2.5 Flare and SunburstGPT Image 2
Image input / cached / output, per 1M tokens$8 / $2 / $30$8 / $2 / $30
Text input / cached, per 1M tokens$5 / $1.25$5 / $1.25
Quality optionslow, medium, high, xhigh, max, auto (default auto)low, medium, high
Default snapshotgpt-image-2.5-flare-2026-09-08, gpt-image-2.5-sunburst-2026-09-08n/a

The same rate does not mean the same bill. The guide warns that "token consumption can differ by model and quality setting," so a Sunburst image at max can cost more than a Flare image at high even though every output token is priced at $30 per million. Use the usage field in the response to measure what a prompt, size, and quality setting consumes, and set explicit quality and size values when you estimate, because auto depends on the generated image. The model pages add that the GPT Image 2 calculator does not estimate GPT Image 2.5 token consumption. (OpenAI image generation guide; GPT Image 2.5 Sunburst model page)

Both models also accept custom dimensions as WIDTHxHEIGHT strings: width and height must be multiples of 16, the aspect ratio must sit between 1:3 and 3:1, no edge may exceed 3,840 pixels, the total pixel count must be between 655,360 and 8,294,400, and resolutions above 2560x1440 are experimental. For transparent backgrounds, set background to transparent and output PNG or WebP. (OpenAI image generation guide)

What To Use When

Choose ChatGPT Images 2.5 when you want a guided creative workflow with Sketch, templates, comments, and the ability to share a prompt. Choose Flare when your product needs fast, high-quality image generation at everyday scale. Choose Sunburst when preserving composition and making precise edits matters more than minimizing generation time. (OpenAI)

For API implementation, the developer guide exposes the models as gpt-image-2.5-flare and gpt-image-2.5-sunburst. The guide also documents image generation and editing through the Images API and the Responses API image-generation tool. (OpenAI image generation guide)

Availability

OpenAI says ChatGPT Images 2.5 is rolling out to all ChatGPT, ChatGPT Work, and Codex users across desktop, mobile, and web. GPT-Image-2.5 Flare and GPT-Image-2.5 Sunburst are available in the API. (OpenAI)

The release is therefore both a ChatGPT feature update and an API model launch. If you are building an image workflow, start with Flare unless your product specifically needs Sunburst's tighter editing control. (OpenAI image generation guide)

Sources

Read next

Related coverage

Frequently Asked Questions

What is ChatGPT Images 2.5?

ChatGPT Images 2.5 is OpenAI's new image model for ChatGPT, with sharper details, faster generation, more reliable editing across multiple turns, and features including Sketch, templates, comments, and shared prompts.

What are GPT-Image-2.5 Flare and Sunburst?

Flare is the faster API model for high-quality everyday image generation and editing. Sunburst is designed for premium workflows where tighter control across edits matters more and longer generation times are acceptable.

Where is ChatGPT Images 2.5 available?

OpenAI says Images 2.5 is rolling out to all ChatGPT, ChatGPT Work, and Codex users across desktop, mobile, and web. The two GPT-Image-2.5 API models are available through the API.

How much do GPT-Image-2.5 Flare and Sunburst cost?

Both models use the same token rates as GPT Image 2: $8 per million image input tokens, $2 cached, $30 per million image output tokens, and $5 per million text input tokens. Equal rates do not mean equal cost per image, because token consumption differs by model and quality setting.

Which GPT-Image-2.5 model should developers choose?

Choose Flare for fast, high-quality everyday generation, high-volume work, or product experiences. Choose Sunburst when detailed editing precision is the priority, such as campaign creative or polished product imagery.

Get the weekly AI Catchup

Tools, practices, and what matters, in your inbox every week.