What Is GPT Image 2.5? OpenAI's New Image Model Explained

OpenAI released ChatGPT Images 2.5 on September 8, 2026, calling it the company’s new state-of-the-art image model. The same generation also lands in the API as two models: GPT-Image-2.5 Flare for everyday speed, and GPT-Image-2.5 Sunburst for higher-precision creative work.

The product page is the source of record: Introducing ChatGPT Images 2.5. OpenAI says people now create more than 3 billion images a week across ChatGPT Images and GPT Image models in the API. Images 2.5 is the follow-up to GPT Image 2, which launched in April 2026.

For PixVerse creators, ChatGPT, the OpenAI API, and PixVerse now all offer the 2.5 generation. On PixVerse, GPT Image 2.5 Flare and GPT Image 2.5 Sunburst sit alongside GPT Image 2, Nano Banana 2, Seedream, Qwen Image, and Kling image models. Generate or edit the still with Flare or Sunburst, then continue into image-to-video in the same workspace. GPT Image 2 remains available when an existing pipeline still calls it. A 14-day launch offer made Flare and Sunburst available at no credit cost to eligible overseas PixVerse members. As of September 27, 2026, do not assume that offer is still active, and do not treat either model as free or unlimited. Check the credit price shown in your workspace. Availability follows the GPT Image 2.5 model page and PixVerse’s September 20, 2026 review.

Key Takeaways

ChatGPT Images 2.5 is rolling out to all ChatGPT, ChatGPT Work, and Codex users across desktop, mobile, and web, including every plan tier.
OpenAI reports up to 50% lower generation latency versus Images 2.0, plus sharper details, more natural lighting, richer textures, and more reliable multi-turn edits.
ChatGPT adds Sketch (@Sketch), format templates such as Poster and Merch, on-image comments, and the option to share the prompt with an image.
The API splits into gpt-image-2.5-flare (fast default) and gpt-image-2.5-sunburst (precision / longer generation). Both add xhigh and max quality settings.
Per-token API rates match GPT Image 2, but OpenAI says the GPT Image 2 cost calculator does not estimate 2.5 token use. Confirm current numbers on OpenAI’s image generation pricing.
Adobe Firefly, Runway, Higgsfield, and Manus are among the first products quoting the new models.

What Changed in Images 2.5?

OpenAI is not positioning this as a new art style so much as a more dependable production model. The launch copy focuses on four practical gains:

Image fidelity. Reference photos keep the original subject more recognizable across new settings, styles, and compositions. Lighting and textures read as more natural.
Precision editing. The model is better at changing only the requested element — a product, a background, a line of copy — while leaving the rest of the frame intact.
Multi-turn consistency. Later edits are more likely to keep earlier decisions, instead of drifting or degrading after several revisions.
Intelligence and layout. Complex instructions, real-world information, requested visual styles, and layouts with transparent backgrounds are described as more reliable.

Those claims come from OpenAI, not from a PixVerse side-by-side lab. Early press matches the same emphasis. TechRadar called the editing tools the biggest day-one change. The Verge used Sketch to turn a rough cat doodle into a photograph, then used a comment to change eye color. Axios, cited in launch coverage, reported stronger partial edits on people, pets, tattoos, and logos.

If you already use GPT Image 2 because it follows long, structured prompts and renders on-image text, Images 2.5 is meant to keep that behavior and make iteration faster. For a spec-level upgrade list, see GPT Image 2.5 vs GPT Image 2.

The ChatGPT product is more than a model swap. OpenAI shipped several creation tools on top of Images 2.5.

Sketch

Type @Sketch in ChatGPT to open a drawing canvas. You sketch a room layout, an outfit contour, or a rough composition, add a style description, and the model treats the doodle as a visual guide. OpenAI’s examples include interior layouts and character outlines. You do not need to draw well; the point is spatial intent that is faster to sketch than to describe.

On mobile, the Help Center flow is: type @, choose Sketch, draw, then continue with written instructions. TechRadar also found Sketch in the Images menu as a dedicated starting option.

Templates

Templates give you a starting format instead of a blank prompt. OpenAI names Poster and Merch. TechRadar’s hands-on also walked through a Product photo template that asks for a product image, then a style (Clean Studio, Editorial, Lifestyle, Dramatic) and a setting before generating. Each template starts with a base prompt and follow-up questions.

That is useful when the job is a known artifact — a flyer, a pack shot, merch mock, icon set — rather than an open-ended illustration.

Comment-based edits

You can now place comments directly on an image. In ChatGPT, those comments are instructions, not notes: click an element, write “remove this” or “change this to blue,” and send. OpenAI says this is meant for focused edits without rewriting the whole prompt.

TechRadar documented an Edit toolbar with Markup, Comment, Remove BG, Erase, and Resize. The same review notes that some of that toolbar appeared during August 2026, so treat the full editor as part of the current ChatGPT Images surface rather than as a guarantee that every control shipped on September 8.

When you share an image, you can include the prompt that produced it. Someone else can rerun the idea with their own photos and details. OpenAI highlighted a viral “what you would have looked like in the ’80s” prompt as an example.

Flare and Sunburst in the API

Developers do not get a single gpt-image-2.5 id. They choose:

API model	Model ID	OpenAI’s role	Snapshot
GPT-Image-2.5 Flare	`gpt-image-2.5-flare`	Fast, high-quality everyday generation; default for most apps	`gpt-image-2.5-flare-2026-09-08`
GPT-Image-2.5 Sunburst	`gpt-image-2.5-sunburst`	Most capable model; tighter control across edits; longer generation	`gpt-image-2.5-sunburst-2026-09-08`

The product announcement says Flare delivers higher-quality images than GPT-Image-2 at 50% lower latency, and is suited to creator and social content, product experiences, visual search, rapid prototyping, and high-volume generation. Sunburst is for production-ready campaign creative and polished product imagery.

The more precise developer guide is slightly more conservative: OpenAI’s GPT Image 2.5 prompting guide calls Flare the small / speed model with quality comparable to GPT Image 2, and Sunburst the base / quality model with higher image quality than GPT Image 2. Both improve precise editing and subject preservation. Use that guide when you are migrating a production pipeline.

Both models:

Accept text and image inputs; output images
Support Image API generations and edits, plus the Responses API image generation tool
Support low, medium, high, xhigh, max, and auto quality
Support opaque and transparent backgrounds (use PNG or WebP for alpha)
Accept custom WIDTHxHEIGHT sizes with the same 4K-class constraints as GPT Image 2

Manus’s evaluation team, quoted by OpenAI, said Flare ran two to four times as fast as GPT-Image-2 in their tests, with better transparent-background generation. That is a customer measurement, not the official 50% latency figure. For a decision table, see GPT Image 2.5 Flare vs Sunburst.

Pricing and Availability

ChatGPT. Images 2.5 is rolling out to ChatGPT, ChatGPT Work, and Codex on desktop, mobile, and web, across all tiers.

API. Standard token rates for both Flare and Sunburst, as listed on the model pages and pricing docs:

Modality	Input	Cached input	Output
Image tokens	$8 / 1M	$2 / 1M	$30 / 1M
Text tokens	$5 / 1M	$1.25 / 1M	not billed (image-only output)

Those rates match GPT Image 2. Actual cost still depends on how many tokens a given size and quality consume. OpenAI explicitly says the GPT Image 2 calculator does not estimate GPT Image 2.5 token consumption, and community reports note no batch pricing for the new models at launch. Measure cost on your own prompts before you lock a budget.

Safety and Provenance

OpenAI says Images 2.5 keeps prompt and image checks, plus C2PA metadata and invisible watermarking to identify images made with its tools. The evaluation write-up is the ChatGPT Images 2.5 system card.

Treat generated people, logos, and brand marks as assets that still need review. GPT Image 2 already struggled with exact logo reproduction in third-party tests; 2.5’s launch materials emphasize subject preservation and local edits, not a claim that proprietary marks now render as vector-perfect.

How GPT Image 2.5 Fits a PixVerse Workflow

PixVerse is an AI video platform with a multi-model image layer. That layer now includes GPT Image 2.5 Flare and GPT Image 2.5 Sunburst alongside GPT Image 2, Nano Banana 2, Seedream, Qwen Image, and Kling image options. The CLI still lists GPT Image 2 as the default image model. Workspace support for Flare and Sunburst does not, by itself, change that CLI default.

A practical 2.5 workflow looks like this:

Generate or edit the still on PixVerse with GPT Image 2.5 Flare for drafts or Sunburst for a locked creative. Use ChatGPT Images 2.5 or the OpenAI API when you need Sketch, comments, or a direct API call.
Keep a video-ready frame: 16:9, 9:16, or 1:1, with the subject large and unobstructed.
Run image-to-video on that still in the same workspace. Upload the file only if it was made outside PixVerse.
If you need character or product continuity across shots, keep the same still as a reference and generate additional angles with 2.5 before animating.

GPT Image 2 on PixVerse remains available when a pipeline is pinned to that model. For prompt patterns that transfer between ChatGPT and a video first frame, use the GPT Image 2.5 prompt guide.

FAQ

Is GPT Image 2.5 the same as ChatGPT Images 2.5?

Almost. ChatGPT Images 2.5 is the consumer product, including Sketch, templates, comments, and sharing. GPT-Image-2.5 Flare and GPT-Image-2.5 Sunburst are the API models. OpenAI’s launch video presents both as the GPT Image 2.5 generation.

Is GPT Image 2.5 available on PixVerse today?

Yes. GPT Image 2.5 Flare and Sunburst are available on PixVerse. Start from a text prompt or a reference image, then use image-to-video when the approved still needs motion. Credit pricing is separate from availability: the 14-day no-credit launch offer should not be treated as current on September 27, 2026. The CLI default remains GPT Image 2 unless the CLI itself lists another default.

Is it faster than GPT Image 2?

OpenAI says generation latency is reduced by up to 50% versus Images 2.0. Flare is the speed-oriented API model. Manus reported 2–4× speed versus GPT-Image-2 in its own evaluations. Latency still depends on size, quality, and whether you pick Flare or Sunburst.

Does it replace Nano Banana 2?

No. Nano Banana 2 remains a strong photorealism and high-volume option on PixVerse. GPT Image models have historically been stronger on structured layouts and on-image text. See GPT Image 2 vs Nano Banana 2 for that comparison; treat 2.5 as an upgrade inside the OpenAI family, not as a new head-to-head we have already scored.