Blog · 2026-11-05

GPT Image 2.6 vs Nano Banana Pro: Metered Tokens or a Flat $0.068

Model pages: GPT Image 2.6 Flare API · GPT Image 2.6 Sunburst API · Gemini 3 Pro Image (Pro) API — pricing, specs and the playground

GPT Image 2.6 is the pick when a job needs animated GIF, SVG or an edit you point at with a markup image or mask, or when you want 1K and 2K images billed by the tokens they use. Nano Banana Pro is the pick when you want one fixed price per image: $0.068 at 1K or 2K and $0.13 at 4K.

The deciding fact is billing. GPT Image 2.6, made by OpenAI, is metered per token, with image output at $30 per 1M tokens. Quality, size, format, reference images and prompt length all move the price. In our tests on 5 November 2026, a 1K low image came to about $0.0077 and a 2K medium image to $0.0516. Nano Banana Pro, built on Google's Gemini 3 Pro, charges the same amount for every image at a given resolution.

GPT Image 2.6 comes in two tiers, Flare and Sunburst. They take the same parameters and bill from the same price list. All three models are called through one apimodels.app endpoint with one API key, so trying the other side means changing one model string.

GPT Image 2.6 vs Nano Banana Pro at a glance

Pricing — GPT Image 2.6 bills image output at $30 per 1M tokens, image input at $8 (cached $2), text input at $5 (cached $1.25) and text output at $10. Each request is rounded up to $0.0001, and SVG is a flat $0.035 per image. Nano Banana Pro costs $0.068 per image at 1K or 2K and $0.13 at 4K.

Resolution and sizes — GPT Image 2.6 takes aspect_ratio with 1K, 2K or 4K, a preset, auto, or an exact WIDTHxHEIGHT. Exact sizes must use multiples of 16, a longest edge up to 3840, a ratio up to 3:1 and 0.66 to 8.29 megapixels. A 16:9 request came back at 1360×768 at 1K and 2720×1536 at 2K. Nano Banana Pro offers 1K, 2K and 4K. Its exact pixel sizes and aspect ratios are not documented.

Inputs — GPT Image 2.6 accepts prompts up to 32,000 characters and up to 16 reference images. It also takes either a markup image or a mask, but not both. Two more controls are reasoning_effort (low, medium or high) and quality, which can be low, medium, high or auto. Nano Banana Pro accepts a text prompt and reference images for editing. Its maximum number of references and mask support are not documented.

Output types — GPT Image 2.6 returns png, jpeg or webp, with transparent backgrounds available. It can also return animated GIF and SVG, and up to 10 images per call. Nano Banana Pro returns images; GIF and SVG output are not documented.

Speed — These are our 5 November 2026 tests of GPT Image 2.6. A 1K low image took about 20 seconds, or about 27 at high quality. A 1K low edit with one reference took 22 to 58 seconds. 2K and 4K images, including high quality, took 13 to 25 seconds. We have no matching test timing for Nano Banana Pro, so speed is not compared here.

Limits and retention — On GPT Image 2.6, SVG is text-to-image only with n set to 1. GIF frames share one sheet of at most 3840 px on the long edge. Both models keep results for 30 days, so download what you need.

Metered tokens against a flat price

Every GPT Image 2.6 cost we measured came in below Nano Banana Pro's flat $0.068 at the same size. All were at 16:9 with one image:

1. 1K low: $0.0077
2. 1K high: $0.0162
3. 1K low edit with one reference: $0.0157
4. 2K low: $0.0283
5. 2K medium: $0.0516

Even at 2K medium the gap is about $0.0164 per image. These are single test calls, not a price list. More reference images, longer prompts, high quality, 4K and GIF all raise the token count. We have not measured a 4K request, so there is no GPT Image 2.6 figure to set against $0.13.

Quality is the setting that moves the meter most easily. apimodels.app defaults GPT Image 2.6 to medium, while the model's own default is high, which uses more tokens. In our tests, passing high roughly doubled the cost of a 1K image, from about $0.0077 to about $0.0162.

The two models also charge your balance differently:

1. GPT Image 2.6 holds an upper bound when a request is accepted. It releases the unused part when the image is ready.
2. Completed GPT Image 2.6 png, jpeg and webp requests return a usage object. Its usage.cost.charged field shows what was actually deducted. GIF, markup or mask edits, SVG and failed tasks return no usage.
3. With Nano Banana Pro, the cost of a batch is the image count times one price, known before you send anything.

Either model can fail on a given call, through moderation refusals or upstream errors. Failed requests are not charged. The account is prepaid, so when the balance runs out, calls return 402 instead of continuing to bill.

GIF, SVG and edits you can point at

Here the two models stop being substitutes. GPT Image 2.6 has output types and controls that are not documented for Nano Banana Pro. If your pipeline depends on one of them, price is no longer the question.

For animation, set output_format to gif. frame_count runs from 2 to 64, with a default of 9. frame_duration_ms runs from 20 to 2000, with a default of 200, so the default clip lasts 1.8 seconds. All frames sit on one sheet whose longest edge can be at most 3840 px. If you pass aspect_ratio with 1K, 2K or 4K, or no size, apimodels.app shrinks the frames to fit. With an exact size you keep the sheet small enough yourself; the whole sheet must stay within 3840 px on the longest edge and 8,294,400 pixels in total.

For vectors, set output_format to svg. It costs a flat $0.035 per image at any quality, works for text-to-image only, and returns one image per request. The file is a static vector with no scripts, external resources or embedded raster. It suits icons, logos and flat illustrations rather than photographic scenes.

For local edits, a markup image is a transparent PNG with circles, arrows or notes drawn over the first reference, and the model edits that reference to match. A mask works differently: white marks the area to change and black the area to keep. In our tests of the previous generation, masks behaved as hints rather than hard boundaries. We have not measured 2.6 yet, so name the region in the prompt too, for example "change only the label on the bottle, keep everything else unchanged".

Nano Banana Pro documents text-to-image and reference-image editing at 1K, 2K and 4K. That covers a product shot, a marketing visual or an edit described in words. It does not cover an animated preview, a logo delivered as a vector file, or an edit steered by arrows on an overlay.

Flare, Sunburst and the 4K question

On the GPT Image 2.6 side you choose a tier, not a price. Flare is the default tier for social posts, catalogue images and high-volume pipelines. Sunburst, as OpenAI positions it, spends more time per image on fine detail. It also keeps edits scoped to the instruction across many rounds. Both tiers bill from the same token price list. Our 5 November 2026 timings fell in the same ranges for both, so we have no measured speed gap between them.

At 4K the comparison turns around. Nano Banana Pro charges a flat $0.13 for a 4K image. GPT Image 2.6 produces 4K too, but its 4K cost is not measured, and 4K uses more tokens than 2K. If you need a fixed budget for 4K hero images, Nano Banana Pro gives you the number in advance. If you want 4K from GPT Image 2.6, send one request, read usage.cost.charged, and price the batch from that.

What a job costs

The GPT Image 2.6 figures below multiply one measured request from 5 November 2026 by the batch size. Treat them as examples, not fixed prices. The Nano Banana Pro figures are exact.

100 catalogue images at 1K

GPT Image 2.6 comes to about $0.77 at 1K low, or about $1.62 at 1K high. Nano Banana Pro costs $6.80.

100 banner images at 2K, 16:9

GPT Image 2.6 comes to about $2.83 at 2K low and about $5.16 at 2K medium. Nano Banana Pro costs $6.80, the same as at 1K.

100 edits with one reference at 1K

GPT Image 2.6 comes to about $1.57 at 1K low. Nano Banana Pro costs $6.80.

10 hero images at 4K, or 100 SVG icons

Ten 4K images on Nano Banana Pro cost $1.30. We have not measured GPT Image 2.6 at 4K and will not guess a figure. For icons, 100 SVG files on GPT Image 2.6 cost exactly $3.50 at the flat $0.035. SVG output is not documented for Nano Banana Pro, so there is no figure to compare.

Which one to use

Pick by the job the image has to do.

1. Choose GPT Image 2.6 Flare for 1K and 2K volume work where the measured per-token costs fit. It is also the pick for anything that needs GIF, SVG, markup or mask edits, exact pixel sizes, or up to 10 images per call.
2. Choose GPT Image 2.6 Sunburst for images that will be inspected up close, such as packaging, print or dense type. It also suits edits revised over several rounds, at the same rates as Flare.
3. Choose Nano Banana Pro when you need a price you can quote before generating: $0.068 at 1K or 2K and $0.13 at 4K, for text-to-image and reference-image editing.

Sometimes neither is the right choice. If you want a GPT Image model with a predictable per-image price, GPT Image 2.5 stays available on apimodels.app with fixed per-image prices. Use 2.6 when you need its new abilities. If you want a flat price at high volume, Nano Banana 2 (Gemini 3.1 Flash Image) costs $0.05 per image at 1K or 2K and $0.08 at 4K on the same endpoint. A photographic scene as a vector is outside both, because SVG on GPT Image 2.6 is meant for icons, logos and flat illustrations.

If Nano Banana Pro already runs a pipeline for you on a fixed per-image budget, do not switch on the figures above alone. Test GPT Image 2.6 on your own prompts and read the charged cost first.

Calling both with one key

Both sides use the same endpoint and the same key. Only the model value and the GPT Image 2.6 fields change.

# GPT Image 2.6 Flare: still image at 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-2.6-flare",
    "prompt": "Ceramic mug on a linen tablecloth, natural light",
    "aspect_ratio": "16:9",
    "quality": "medium",
    "reasoning_effort": "low"
  }'

# GPT Image 2.6 Sunburst: animated GIF, 9 frames of 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-sunburst",
    "prompt": "A paper boat rocking on small waves, flat illustration",
    "aspect_ratio": "16:9",
    "output_format": "gif",
    "frame_count": 9,
    "frame_duration_ms": 200
  }'

# Nano Banana Pro: flat price per image
curl https://api.apimodels.app/v1/images/generations \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "nanobananapro",
    "prompt": "Ceramic mug on a linen tablecloth, natural light"
  }'

# Each POST returns a taskId; poll it with GET /v1/images/generations?task_id=...
curl "https://api.apimodels.app/v1/images/generations?task_id=YOUR_TASK_ID" \
  -H "Authorization: Bearer YOUR_API_KEY"

FAQ