Blog · 2026-11-05

GPT Image 2.6 vs GPT Image 2: Token-Billed Control or a Flat Price

Model pages: GPT Image 2.6 Flare API · GPT Image 2.6 Sunburst API · GPT Image 2 API — pricing, specs and the playground

GPT Image 2 is the right call when you want a known price per image and nothing to tune; GPT Image 2.6 is the right call when you need animated GIF or SVG output, markup-image or mask edits, exact pixel sizes, or control over quality and reasoning effort.

The deciding difference is billing. On apimodels.app, GPT Image 2 costs $0.025, $0.03 or $0.05 per image at 1K, 2K or 4K and always runs the medium quality tier, so the price is known before you send anything. GPT Image 2.6 is billed per token, and quality, size, reference images, GIF frames and the prompt all change the count. In our 5 November 2026 tests a 1K low image came to about $0.0077 and a 2K medium image to $0.0516, so 2.6 costs less than GPT Image 2 on some jobs and more on others.

GPT Image 2.6 comes from OpenAI in two tiers, Flare (gpt-image-2.6-flare) and Sunburst (gpt-image-2.6-sunburst). They take the same parameters and bill from the same price list, so what follows applies to both unless a tier is named.

GPT Image 2.6 vs GPT Image 2 at a glance

Pricing — GPT Image 2 charges $0.025 / $0.03 / $0.05 per image at 1K / 2K / 4K, edits included. GPT Image 2.6 charges $30 per 1M image output tokens, $8 per 1M image input tokens (cached $2), $5 per 1M text input tokens (cached $1.25) and $10 per 1M text output tokens, each request rounded up to $0.0001; SVG is a flat $0.035 per image.

Resolution and sizes — GPT Image 2 renders at 1K, 2K or 4K with aspect ratios 1:1, 2:3, 3:2, 4:5, 5:4, 4:3, 3:4, 16:9, 9:16 and 21:9, with 5:4 and 4:5 at 1K only. GPT Image 2.6 takes an aspect ratio with 1K, 2K or 4K, a preset name, auto, or an exact WIDTHxHEIGHT: multiples of 16, longest edge up to 3840, ratio up to 3:1, 0.66 to 8.29 megapixels.

Inputs — Both accept up to 16 reference images per request. GPT Image 2.6 also accepts a markup image, a transparent PNG with circles, arrows or notes drawn over the first reference, or a mask where white marks the area to change and black the area to keep, but not both at once. Neither markup images nor masks are documented for GPT Image 2.

Outputs — GPT Image 2 documents text-to-image and image editing. GPT Image 2.6 adds animated GIF with 2 to 64 frames, SVG from a text prompt, transparent backgrounds and up to 10 images per call.

Quality controls — GPT Image 2 always runs medium, and a quality field has no effect. GPT Image 2.6 takes quality low, medium, high or auto, defaulting to medium on apimodels.app (the model's own default is high, which uses more tokens), plus reasoning_effort (low, medium or high), which trades latency for detail.

Speed — In our 5 November 2026 tests, GPT Image 2.6 took about 20 seconds for a 1K low text-to-image (about 27 at high quality), 22 to 58 seconds for a 1K low edit with one reference, and 13 to 25 seconds at 2K and 4K, high quality included, with the same ranges for both tiers. GPT Image 2 is documented at a median of 60 seconds over 30 days in production, with 95% of requests back within 157 seconds; set client timeouts above 180 seconds. Test calls and production traffic are different measurements, so read the gap as a direction.

Limits and retention — On GPT Image 2.6, SVG is text-to-image only with one image per request, and GIF frames share one sheet of at most 3840 px on the long edge. On GPT Image 2, 4K is capped at 8,294,400 pixels, so a 1:1 image at 4K is 2880x2880. Results from both are kept for 30 days.

A fixed price or a per-token price

GPT Image 2 settles the budget before you write any code: a 1K image is $0.025 whether it is a fresh prompt or an edit with several references, because quality is pinned to medium and price follows resolution alone.

GPT Image 2.6 meters what each request uses. Our measured 16:9 single-image requests came to about $0.0077 at 1K low, $0.0162 at 1K high and $0.0157 for a 1K low edit with one reference, all below GPT Image 2's $0.025. At 2K the result splits: $0.0283 at low quality sits below GPT Image 2's $0.03, while $0.0516 at medium sits above it. Each extra reference adds image input tokens at $8 per 1M, a long prompt adds text tokens, and higher quality or GIF adds output tokens, so these figures are examples, not a price list.

Two mechanisms make per-token billing easier to run. When a 2.6 request is accepted, an upper bound is held and the unused part is released once the image is ready. Completed png, jpeg and webp text-to-image and edit requests also return a usage object in the OpenAI Images shape, with usage.cost split into amount, discount and charged; charged is what was actually deducted and the field to log. GIF, markup or mask edits and SVG return no usage object.

If finance wants one per-image figure in a spreadsheet, GPT Image 2 gives you that; if you would rather pay for exactly the quality and size each request uses, 2.6 does.

What 2.6 produces that GPT Image 2 does not document

Set output_format to gif on GPT Image 2.6 and you get an animation of 2 to 64 frames (9 by default), each shown for 20 to 2000 ms (200 by default). All frames are laid out on one sheet whose longest edge can be at most 3840 px. Pass an aspect ratio with 1K, 2K or 4K, or no size, and apimodels.app shrinks the frames to fit; with an exact size you keep it small enough yourself, and the whole sheet must stay within 3840 px on the longest edge and 8,294,400 pixels in total.

SVG comes from a text prompt only, one image per request (n must be 1), at a flat $0.035 whatever the quality setting. The file is a static vector with no scripts, external resources or embedded raster, so it suits icons, logos and flat illustrations rather than photographic scenes.

Editing gains a way to point. A markup image lets you draw circles, arrows or notes over the first reference, and the model edits it accordingly; a mask marks the area to change in white. We have not measured how strictly 2.6 respects masks, and on the previous generation they behaved as hints rather than hard boundaries, so name the region in the prompt too, for example "change only the label on the bottle, keep everything else unchanged".

GPT Image 2's page lists text-to-image and editing with up to 16 reference images. GIF, SVG, markup images and masks are not documented for it, so if any of them is in your spec, the choice is made.

Sizes and quality: one setting or a dial

GPT Image 2 works from a resolution tier plus an aspect ratio, and the exact pixels can vary within a tier. A 2K request at 16:9 comes back at 2048x1152 or 2736x1536, both billed at $0.03, and every completed task reports the delivered width and height. That suits most layouts but is awkward when a template expects one exact size.

GPT Image 2.6 accepts an exact WIDTHxHEIGHT within the limits above; in our tests 16:9 requests were delivered at 1360×768 for 1K and 2720×1536 for 2K. It also hands you the two settings GPT Image 2 fixes: quality, and reasoning_effort for more detail at the cost of latency. Because apimodels.app defaults 2.6 to medium, leaving quality out does not quietly bill at high.

What a job costs

GPT Image 2 figures below are fixed prices. GPT Image 2.6 figures multiply our measured 16:9 single-image requests from 5 November 2026 by the job size, so they are examples; your token counts will move with prompt length, references and aspect ratio.

1. 100 product shots at 1K: GPT Image 2 costs $2.50 at medium. GPT Image 2.6 comes to about $0.77 at low or $1.62 at high; 1K medium was not priced in our tests.
2. 100 single-reference edits at 1K: GPT Image 2 costs $2.50. GPT Image 2.6 comes to about $1.57 at 1K low, with more image input tokens for every additional reference.
3. 100 banners at 2K, 16:9: GPT Image 2 costs $3.00. GPT Image 2.6 comes to about $2.83 at low or $5.16 at medium.
4. 100 SVG icons: GPT Image 2.6 costs $3.50 flat. GPT Image 2 documents no SVG output.

For one 4K hero image, GPT Image 2 costs $0.05. We have no measured 4K price for 2.6, so check usage.cost.charged on your first calls before budgeting. On both models, generation can fail through moderation refusals or upstream errors, and failed requests are not charged.

Which one to use

Pick GPT Image 2 if you price work per image, medium quality is enough, your outputs are still images and your layouts fit its aspect ratios, 21:9 included. It is also the simpler choice for a pipeline already budgeted at $0.025 / $0.03 / $0.05.

Pick GPT Image 2.6 Flare if you need GIF, SVG, markup or mask edits, exact pixel sizes, low-quality 1K drafts or up to 10 images per call. It is the default 2.6 tier for everyday volume.

Pick GPT Image 2.6 Sunburst when someone will look closely: packaging, print, product detail, dense type, or edits that must change only what you asked across several rounds. OpenAI positions it as the tier that spends more time per image on detail; its per-token prices are the same as Flare's, so the tier does not change the bill.

One case where neither fits: if you want to choose the quality tier but keep a fixed per-image price, gpt-image-2-all is the same underlying model as GPT Image 2 with low, medium, high and auto exposed, from $0.01 at 1K low. And if GPT Image 2 already does the job at medium and you need none of 2.6's new outputs, a lower measured 1K price is not on its own a reason to give up a price you can predict.

Calling both with one key

All three models use the same endpoint and the same API key, so switching is a change to the model field plus any 2.6-only fields.

# GPT Image 2: flat price by resolution, always medium quality
curl https://api.apimodels.app/v1/images/generations \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-image-2",
    "prompt": "A ceramic mug on a linen tablecloth, soft morning light",
    "aspect_ratio": "16:9"
  }'

# GPT Image 2.6 Flare: animated GIF, 9 frames at 200 ms each
curl https://api.apimodels.app/v1/images/generations \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-image-2.6-flare",
    "prompt": "A paper boat drifting across a puddle, flat illustration",
    "aspect_ratio": "1:1",
    "output_format": "gif",
    "frame_count": 9,
    "frame_duration_ms": 200
  }'

# GPT Image 2.6 Sunburst: billed per token, quality and reasoning effort set per request
curl https://api.apimodels.app/v1/images/generations \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-image-2.6-sunburst",
    "prompt": "Close-up of a perfume bottle label with fine embossed type",
    "aspect_ratio": "16:9",
    "quality": "high",
    "reasoning_effort": "low"
  }'

# Each call returns a taskId; poll GET /v1/images/generations?task_id=... until the image is ready

FAQ