Blog · 2026-11-05

GPT Image 2.6 vs Nano Banana 2: Per-Token Control or a Flat $0.05

Model pages: GPT Image 2.6 Flare API · GPT Image 2.6 Sunburst API · Nanobanana2 API — pricing, specs and the playground

Choose Nano Banana 2 when you want a known price per image, $0.05 at 1K or 2K and $0.08 at 4K, for text-to-image and edits; choose GPT Image 2.6 when the job needs animated GIF, SVG, markup-guided edits, masks or up to 16 reference images. The deciding fact is billing: Nano Banana 2 charges per image, while GPT Image 2.6 charges per token, with image output at $30 per 1M tokens, so its price moves with quality, size, format and references.

Per-token billing does not mean a higher bill. In our 5 November 2026 tests a 1K low request on GPT Image 2.6 came to about $0.0077 and a 2K medium request to $0.0516, both at 16:9 with one image. At low quality GPT Image 2.6 costs less than Nano Banana 2's flat $0.05; at 2K medium the two sit within a fraction of a cent. What tokens take away is knowing the exact cost before you send the request.

GPT Image 2.6, made by OpenAI, comes in two tiers, Flare and Sunburst, on one token price list. Nano Banana 2 is Google's model, built on Gemini 3.1 Flash. On apimodels.app all three share one endpoint and one key.

GPT Image 2.6 vs Nano Banana 2 at a glance

Pricing — GPT Image 2.6: image output $30 per 1M tokens, image input $8 (cached $2), text input $5 (cached $1.25) and text output $10, each request rounded up to $0.0001, the same rates as OpenAI's official API. SVG is the exception at a flat $0.035 per image. Nano Banana 2: $0.05 per image at 1K and 2K and $0.08 at 4K, for text-to-image and edits alike.

Resolution and sizes — GPT Image 2.6 accepts aspect_ratio with 1K, 2K or 4K, an exact WIDTHxHEIGHT (multiples of 16, longest edge up to 3840, ratio up to 3:1, 0.66 to 8.29 megapixels), a preset name or auto. In our tests a 1K 16:9 request delivered 1360×768 and a 2K 16:9 request 2720×1536. Nano Banana 2 offers 1K, 2K and 4K; exact pixel sizes are not documented for it.

Inputs — GPT Image 2.6 takes a prompt of up to 32,000 characters, up to 16 reference images, and either a markup image or a mask to guide an edit, never both, plus quality (low, medium, high or auto) and reasoning_effort (low, medium or high), which trades latency for detail. Nano Banana 2 takes a prompt and reference images for natural-language editing; its reference limit and any mask support are not documented.

Output types — GPT Image 2.6 returns png, jpeg or webp, animated GIF and SVG, up to 10 images per call, with transparent backgrounds listed for Flare. Nano Banana 2 is listed for images at 1K, 2K or 4K; GIF, SVG, transparency and multiple outputs per call are not documented.

Speed — For GPT Image 2.6 our 5 November 2026 tests measured about 20 seconds for a 1K low text-to-image, about 27 seconds at high quality, 22 to 58 seconds for a 1K low edit with one reference, and 13 to 25 seconds for 2K and 4K images, high quality included. Nano Banana 2 is described as tuned for quick output, but we have no measured timing to quote.

Limits — GPT Image 2.6's SVG is text-to-image only, one image per request (n must be 1), and quality has no effect on it; GIF frames share one sheet of at most 3840 px on the longest edge. Its results are kept for 30 days; the Nano Banana 2 listing does not state a retention period. Either model can fail on a given call through a moderation refusal or an upstream error, and failed requests are not charged.

A flat price against a token meter

With Nano Banana 2 the cost of a batch is known before the first request: count the images and multiply. 1,000 images at 2K cost $50 whatever the prompt says, and a 4K image is $0.08 every time.

GPT Image 2.6 meters each part of a request. The prompt is text input at $5 per 1M tokens, each reference image is image input at $8 per 1M ($2 when cached), and the image you get back is image output at $30 per 1M. Quality, size, GIF output, references and prompt length all change the token count, so the same 16:9 prompt came to about $0.0077 at 1K low and about $0.0162 at 1K high in our tests. Treat those as measured examples, not a price list.

Two mechanics keep the meter manageable. When a request is accepted, apimodels.app holds an upper bound against your balance and releases the unused part when the image is ready, so a large batch needs headroom above its expected cost. Completed png, jpeg and webp requests also return a usage object in the OpenAI Images shape, where usage.cost carries amount, discount and charged; charged is what was deducted and the figure to log. GIF, markup or mask edits and SVG come back without usage, so those calls have no token breakdown.

Check one default in ported code: apimodels.app runs GPT Image 2.6 at medium quality unless told otherwise, while the model's own default is high, which uses more tokens on every image.

What GPT Image 2.6 does that Nano Banana 2 does not document

This is where the two are built for different jobs. Nano Banana 2 is positioned for high-volume, cost-sensitive generation such as social visuals, rapid iteration and batched creative, through text-to-image or reference-image edits. GPT Image 2.6 adds output types and edit controls that change what one API call can produce.

The additions that matter for a build decision:

1. Animated GIF: set output_format to gif, with frame_count from 2 to 64 (default 9) and frame_duration_ms from 20 to 2000 (default 200).
2. SVG: set output_format to svg for a static vector from a text prompt, at a flat $0.035 per image, whatever the quality.
3. Markup-image edits: draw circles, arrows or notes on a transparent PNG and the model edits the first reference to match.
4. Masks: white marks the area to change and black the area to keep, as the alternative to a markup image.
5. Batches: up to 10 images out and up to 16 references in per call.

Each comes with a condition. The SVG file has no scripts, external resources or embedded raster, so it suits icons, logos and flat illustrations rather than photographic scenes; 100 icons cost $3.50. For GIF, all frames sit on one sheet of at most 3840 px on the long edge: with aspect_ratio plus 1K, 2K or 4K, apimodels.app shrinks the frames so the sheet fits, and with an exact size you keep it small yourself, so that the whole sheet stays within 3840 px on the longest edge and 8,294,400 pixels in total. We have not measured how strictly 2.6 respects masks; on the previous generation they behaved as hints rather than hard boundaries, so name the region in the prompt as well.

If the job is a logo set, an animated sticker or a product photo where only the label should change, Nano Banana 2's documented features do not cover it, and price stops being the question.

Flare, Sunburst and where Nano Banana 2 fits

The two GPT Image 2.6 tiers take the same parameters and bill from the same token list, so the tier does not change the rate; what you request does. Flare is the default tier for everyday volume: social posts, catalogue images, pipelines. Sunburst spends more time per image on fine detail such as fabric weave, fine type and reflective surfaces, and its edits are meant to hold subject and composition across many rounds of revision. Our measured timings are the same ranges for both tiers, so speed is not the reason to choose one.

Nano Banana 2 has its own step up on the same endpoint: Nano Banana Pro, on Gemini 3 Pro Image, is positioned for maximum quality, and you can switch between the two freely. Its pricing is outside this comparison.

What a job costs

GPT Image 2.6 figures below come from our measured requests on 5 November 2026 at 16:9 with one image per request. They are examples of what those settings cost, not fixed prices; a longer prompt or extra references will move them. Nano Banana 2 figures are its listed per-image prices.

100 social images at 1K — Nano Banana 2: $5. GPT Image 2.6: about $0.77 at 1K low, about $1.62 at 1K high.

100 edits at 1K, one reference each — Nano Banana 2: $5. GPT Image 2.6: about $1.57 at 1K low.

1,000 catalogue images at 2K — Nano Banana 2: $50. GPT Image 2.6: about $28.30 at 2K low and about $51.60 at 2K medium, which is $1.60 above Nano Banana 2 across the batch, or $0.0016 per image.

One 4K hero image — Nano Banana 2: $0.08. GPT Image 2.6 at 4K has no measured cost in our tests; 4K and high quality use more tokens, so run one request and read its usage before pricing a batch.

Which one to use

Pick Nano Banana 2 for high-volume text-to-image or reference edits at 1K, 2K or 4K when every request should be priced before it is sent, especially at 4K, where its $0.08 is fixed and GPT Image 2.6's cost is unmeasured.

Pick GPT Image 2.6 Flare when you need GIF, SVG, markup or mask edits, exact pixel sizes or several images per call, or when you generate at 1K low in volume, where our measured cost sat below Nano Banana 2's flat price. Pick Sunburst for the same features when the image will be inspected up close, such as packaging, print or dense type, or when an edit has to hold across many rounds.

Do not move a working Nano Banana 2 pipeline that renders 2K images to GPT Image 2.6 at medium quality for price alone: our 2K medium request came to $0.0516 against $0.05, so you would swap a fixed price for a meter and save nothing. And if you want OpenAI's model at a fixed price per image without the 2.6 additions, GPT Image 2.5 remains on apimodels.app with fixed per-image prices.

Calling both with one key

The model string is the only change for a basic call; the quality, reasoning and GIF fields apply to GPT Image 2.6 only.

# GPT Image 2.6 Flare: text-to-image, 16:9, low 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": "A ceramic mug on a wooden table in morning light", "aspect_ratio": "16:9", "quality": "low", "reasoning_effort": "low"}'

# GPT Image 2.6 Sunburst: 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-sunburst", "prompt": "A paper boat bobbing on small waves", "aspect_ratio": "16:9", "output_format": "gif", "frame_count": 9, "frame_duration_ms": 200}'

# Nano Banana 2: text-to-image, 16:9
curl https://api.apimodels.app/v1/images/generations \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "nanobanana2", "prompt": "A ceramic mug on a wooden table in morning light", "aspect_ratio": "16:9"}'

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

FAQ