Compare available image models in iMini

Pay for the workflow that produces an acceptable asset, not the largest version number. A model that makes an attractive first draft can still be expensive if you spend five more attempts correcting a product label, a face or a layout.

GPT Image 2.6 is documented by OpenAI, with Sunburst for detailed visual work and Flare for fast iteration. GPT Image 2.5 also has Sunburst and Flare variants; GPT Image 2 is the earlier model. This guide compares their documented options and gives you a buying test, rather than claiming an unmeasured quality winner.

Which version should you choose?

Evaluate 2.6 when you need its documented complexity control or SVG and animated GIF generation. Keep 2.5 or 2 in consideration when your existing raster-image workflow already meets the brief. Within the newer families, shortlist Sunburst for detail-sensitive editing and Flare for iteration, then measure the results on your own inputs. Official positioning is a starting point, not a measured speed or accuracy guarantee.

What changes the purchase decision?

2.6 adds controls, not a guarantee of a better result

The complexity setting changes planning before rendering; quality controls rendering quality separately. OpenAI documents low, medium and high complexity, with high as the default. Higher complexity can help with dense text or object relationships, but can also add latency and unwanted changes. For a focused background edit, test lower complexity instead of assuming high is always best.

SVG and animated GIF output are documented for the generations endpoint, not the image-edit endpoint. SVG production includes a reference image and an authoring step, with both contributing to cost. If you need a photographic product edit, these export options may add little value to your purchase. Markup editing is also documented for supported 2.5 models, so it should not be presented as exclusive to 2.6.

GIF output changes what you should test before paying

A short product loop can help demonstrate a product or develop an ad concept, but a polished opening frame is only the starting point. For a perfume-bottle animation, include label readability, bottle shape and background stability in the test brief; for a waterfall, specify which elements should move and the required frame count and aspect ratio.

The buying lesson is to inspect every frame, including the transition back to the start. Check labels, silhouettes, hands, shadows and objects that should stay still, then verify the downloaded file dimensions, frame count and timing. A motion request that looks plausible can still fail the delivery brief.

OpenAI documents GIF generation for 2.6 through the Image API generation endpoint. Its lantern example uses nine frames at 200 milliseconds per frame for a 1.8-second cycle. The size setting describes each finished frame, and frame size, count and number of animations share a pixel budget. A small sample is not the model's resolution ceiling; static-image 4K limits do not establish a GIF limit. GIF has no audio, and app controls can differ from the API.

Official Sunburst GIF example with nine frames at 200 milliseconds each; this is not an iMini test result. OpenAI

SVG is an upgrade reason when you need editable vector graphics

For an icon, illustration or logo concept that must work at several sizes, editable vector paths can matter more than photographic detail. OpenAI's SVG workflow generates a reference image and then produces static vector content with editable shapes and colors. It does not turn any photograph into a layered Photoshop document or generate SVG animation.

Inspect the actual SVG in your design tool, not just its preview. Check text, paths, colors and scaling before counting it as a usable asset. SVG output is documented for the generation endpoint with one result per request, and the reference-image and SVG-authoring steps both contribute to cost. For a standard product-photo edit, this feature alone may not justify switching.

Official Flare SVG example rendered as an image; inspect the original vector file before delivery. OpenAI

Add a delivery check to your model comparison

A prompt for testing a simple product loop

Create a short looping GIF of a fictional amber glass perfume bottle labeled SOL on a studio surface. Use a locked camera and one small side-to-side rotation, returning to the opening pose. Keep the label, cap, bottle shape and background consistent. Add no text or sound. Request nine frames at 200 milliseconds per frame, then verify the exported file.

Use this prompt as a test brief, not a promise of exact output; set supported export controls explicitly and confirm current app availability before paying.

Price per token is not price per finished image

The official standard API rates checked for this guide are the same across the three families shown below. They are USD per million tokens, not a fixed per-picture charge. Input references, output token consumption, size, quality and repeated attempts all affect the bill. The Responses API can also add mainline-model usage.

A ChatGPT subscription, direct OpenAI API usage and an iMini plan are different purchases. An app can charge credits, restrict models or expose fewer settings. Before subscribing, check the exact model, credit cost for your size and quality, download options, renewal terms and how failed attempts are handled. API pricing does not establish any app’s retail price.

Official standard API token rates

These are standard uncached rates checked on November 5, 2026. Cached, Batch and interface-specific billing conditions differ. Equal rates do not imply equal token consumption, equal image cost or equal subscription value. Recheck the official pricing page before committing a budget.

Run a small buying test on work you actually need

Use one product edit, one text-heavy promotion and one layout-sensitive image. Keep the source, prompt, dimensions and common quality setting the same. Run several attempts within a fixed budget and keep failures in the record. If a model lacks a needed setting, record that limitation instead of silently changing the task.

Count accepted assets, not attractive samples. Reject wrong labels, altered products, changed faces or unreadable offer text before counting a result as usable. Record elapsed time and cleanup time as well as charges. This is a test plan, not a claim that we have benchmarked these models.

When is an upgrade worth paying for?

Use total generation charges divided by accepted assets as your generation cost per usable image; separately add your cleanup time and any plan fee. For illustration only, a $3 test with six accepted assets costs $0.50 per accepted asset; a $2 test with two accepted assets costs $1. These are hypothetical budgets, not model prices or benchmark results.

Upgrade when a necessary export or control is available, or your measured reduction in retries and cleanup justifies the price. Stay with your current model when it meets the brief reliably. For repeated campaign work, the best model may differ between rough concepts and final product edits.

How to use this comparison in iMini

Open the image workspace, attach the same reference where supported and compare models that are actually available. Review the current pricing page before purchasing. Save the model name, settings and prompt with each accepted result so you can reproduce the workflow later.

Check current iMini plans

Two prompts for a fair comparison

Product-edit test

Edit the uploaded product photo. Change only the background to a warm neutral studio scene with a soft contact shadow. Retain the product silhouette, proportions, cap, material and every visible label character. Keep the camera angle and product placement. No added packaging, props or text.

Text-and-layout test

Create a square promotional graphic for a fictional ceramic mug. Exact headline: WINTER EDIT. Exact secondary text: Handmade ceramic mug. Use one mug on the right and a clear left text column with generous margins. Add no price, discount, logo or other wording. Keep all lettering legible.

Avoid a misleading comparison

Do not compare one model at low quality with another at max, pick only the best sample, or count a prettier but incorrect product as a win. Do not buy access on the assumption that “2.6” includes every API feature or that “2” must be cheaper. Check exact names and controls, then choose by accepted output and total effort.

Frequently asked questions

Is GPT Image 2.6 officially documented?

Yes. OpenAI’s current image-generation guide and model pages document GPT Image 2.6 Sunburst and Flare. Account and app access still need checking.

Is GPT Image 2.6 cheaper than 2.5?

Not on the standard per-token rates listed here. Actual cost per accepted image can differ with usage, settings and retries.

Should I buy Sunburst or Flare?

Start from detailed editing versus fast iteration, then test the same brief. No measured winner is claimed here.

Can GPT Image 2.6 produce SVG and animated GIF files?

The official Image API documents those formats for generation, with specific constraints. They are not supported as outputs of the image-edit endpoint in that guide.

Is upgrading always better for product photos?

No. Label fidelity, product shape, usable output rate and cleanup effort matter more than the version number alone.

Continue with a practical workflow

A closer look at 2.5 Flare and Sunburst · Photo editing prompts and follow-up edits · Background editing with product checks

Official sources

Features and standard API rates were checked against the official OpenAI documentation on November 5, 2026. This is a documented comparison and buying framework, not a hands-on performance ranking.

OpenAI Image generation · OpenAI API pricing · GPT Image 2.6 Sunburst

Choose an image model for your next real task