Is There a Nano Banana 3? Status, Models and API (2026)

Is Nano Banana 3 out yet?

Nano banana 3 is not a released model as of late October 2026. Google has not shipped anything under that name. What does exist is a set of reports from mid-October 2026 saying Google appears to be preparing a new image generation model, likely to be called Nano Banana 3 Pro. Until Google publishes it, any site offering you “Nano Banana 3” access is selling either a rumour or an existing model with a new label stuck on it.

So why does the search keep growing? Three reasons explain most of it:

• Nano Banana Pro is built on Gemini 3. Its official name is Gemini 3 Pro Image. Because of the “3”, many people call it nano banana 3, and some coverage around its launch used that phrase loosely.
• The version numbers in the family do not match Gemini numbers. Nano Banana 2 is officially Gemini 3.1 Flash Image. A “2” sitting on a Gemini 3.x model makes people assume a “3” must already exist somewhere.
• The reports of a coming Nano Banana 3 Pro. Once a name leaks, people search for it right away, long before there is anything to use.

This article sorts out the names, explains what the current models can actually do, gives you the per-image cost of each one, and shows how to write your code so that moving to nano banana 3 on release day is a one-line change rather than a rewrite.

The Google Nano Banana family, decoded

“Nano Banana” began as a codename. Google’s image model appeared under that playful name on an anonymous public comparison leaderboard, where people vote between images from unlabelled models. When Google confirmed it as Gemini 2.5 Flash Image, the nickname had already stuck, and Google leaned into it for later releases.

Two terms help before you read the table. Flash is Google’s name for its faster, cheaper model tier. Pro is the higher-quality tier that takes more compute per request and costs more. Most of the Nano Banana family is some mix of a Gemini version number and one of these tiers.

Popular name	Official name	Tier	Unicorn API model id
Nano Banana	Gemini 2.5 Flash Image	Fast, natural-looking images and edits	nano-banana
Nano Banana Pro	Gemini 3 Pro Image	High fidelity, 1K to 4K	nano-banana-pro
Nano Banana 2	Gemini 3.1 Flash Image	Flash tier	See the model catalog
Nano Banana 2 Lite	Lite version of Nano Banana 2	Cheapest, fast 1K images	nano-banana-2-lite
Nano Banana 3 / 3 Pro	Not released	Reported only	None yet

A few common points of confusion, answered directly:

• “Nano banana 2.0” is simply how some people write Nano Banana 2. There is no separate 2.0 model.
• Nano Banana 2 and Nano Banana Pro are not the same thing. Pro is the high-end Gemini 3 Pro Image model. Nano Banana 2 is a Flash-tier model built on Gemini 3.1. The higher number does not mean higher quality across the board; it means a different tier on a newer base.
• “Gemini nano banana” and “google nano banana” refer to the same family. Gemini is the model line, Nano Banana is the image-model nickname.
• Gemini Nano is unrelated. It is Google’s small language model built to run on phones. The “nano” in Nano Banana is part of a codename, not a description of size.

What Nano Banana Pro does today

If what you really want is “the newest, best Google image model”, that is Nano Banana Pro, and it is the model anyone searching for nano banana 3 should test first. Here is what it offers through Unicorn API.

Accurate text inside images

Text rendering, meaning readable words drawn as part of the image, used to be the weak spot of image models. Signs, labels and posters came out as near-letters. Nano Banana Pro is built to render text accurately, which makes it a practical choice for posters, menus, packaging mockups, infographics and thumbnails with titles.

1K, 2K or 4K output

You choose the resolution per request with the resolution field: 1K, 2K or 4K. 4K is useful when the image will be printed, cropped heavily or shown on large screens. For web use, 2K is usually plenty, and on Unicorn API it costs the same as 1K.

Up to 8 reference images

Reference images are pictures you send alongside the prompt so the model can copy a product, a person, a style or a layout. Nano Banana Pro accepts up to 8 through the ref_images field. That is enough to give it a product from several angles plus a style reference, which helps keep a character or item consistent across a series.

Aspect ratios

Supported ratios are 16:9, 1:1, 9:16, 21:9 and 4:3. That covers widescreen banners, square social posts, vertical stories and ultrawide headers.

How it compares with its siblings

• Nano Banana gives quick, natural-looking images and edits with up to 3 reference images. It sets size with an image_size field rather than resolution.
• Nano Banana 2 Lite is the cheapest Gemini image model: fast 1K images and edits with up to 10 reference images, and two extra ratios, 3:2 and 2:3. It has no resolution option.

The practical summary: Pro for quality and text, Lite for volume and many references, the original Nano Banana for quick natural edits.

What to expect from nano banana 3, and what to ignore

Nobody outside Google knows the specifications of nano banana 3, so be suspicious of any page that lists them confidently. What you can reasonably say is limited:

• The reported name is “Nano Banana 3 Pro”. If that holds, the first release would be the high-end tier, a successor to Nano Banana Pro, rather than a cheap Flash model.
• Google’s pattern so far is Pro and Flash tiers. The family already has both a Pro model and Flash models with a Lite variant. A new generation may follow the same split, but that is a guess, not a fact.
• It will be a cloud model. Every Nano Banana model runs on Google’s servers and is reached through apps or an API.

And here is what to ignore:

• “Nano Banana 3 download” pages. There has never been a downloadable Nano Banana model. An installer claiming to be one is at best useless and at worst malware.
• Benchmark charts for an unreleased model. If the model is not out, the numbers are made up or come from a different model.
• Claims of 3D output. Nano Banana models produce 2D images. You can prompt for isometric renders, turnarounds or clay-style figures and feed those into a separate 3D tool, but the output is a flat image file.
• Reddit screenshots labelled “nano banana 3”. Plenty of these are Nano Banana Pro outputs relabelled. A screenshot proves nothing about which model made it.

When the model does arrive, judge it on your own work. Keep a fixed set of 10 to 20 prompts that represent what you actually make, including your hardest cases: small text, hands, product labels, a specific brand colour. Run them through the new model with the same settings you use today, then compare side by side. That tells you more than any review.

Nano Banana prices per image in credits

Unicorn API prices every request in credits. On a monthly plan, credits cost $0.01 each (Starter $25, Creator $50, Studio $100, Pro $200 per month). Top-up credits cost $0.015 each and never expire. Prices below are per output image; when you set n to request several images, the price is multiplied by n. Full details are on the pricing page.

Model and setting	Credits per image	On plan credits	On top-up credits
Nano Banana Pro, 1K	18	$0.18	$0.27
Nano Banana Pro, 2K	18	$0.18	$0.27
Nano Banana Pro, 4K	24	$0.24	$0.36
Nano Banana	4	$0.04	$0.06
Nano Banana 2 Lite	4	$0.04	$0.06
GPT Image 1.5, low / medium / high	3 / 4 / 22	$0.03 / $0.04 / $0.22	$0.045 / $0.06 / $0.33
Flux 3 Klein Text to Image, 1K / 2K / 4K	2 / 3 / 7 (one image)	$0.02 / $0.03 / $0.07	$0.03 / $0.045 / $0.105

The Flux 3 Klein models are listed at 1.6, 2.5 and 6.5 credits per image, but totals are rounded up to whole credits after multiplying by the number of images. One 1K image therefore costs 2 credits, while four 1K images cost 6.4 rounded up to 7. The same pricing applies to Flux 3 Klein Edit Image. If you batch, request several images in one call to waste less to rounding.

A worked example

Say you need 10 finished product images for a launch page. A sensible workflow:

1. Explore 40 drafts with Nano Banana 2 Lite: 40 × 4 = 160 credits.
2. Pick the 10 best directions and render them with Nano Banana Pro at 4K: 10 × 24 = 240 credits.
3. Total: 400 credits, which is $4.00 on plan credits or $6.00 on top-up credits.

Running all 50 images on Nano Banana Pro at 4K would cost 1,200 credits instead, three times as much, for drafts you would throw away. For a fuller breakdown across the family, see our Nano Banana pricing guide. You can also get the exact price of any request before running it with POST /v1/models/price, documented in the API docs.

Which Nano Banana model should you use right now?

“Best” depends on the job. Use this as a starting point:

• Posters, thumbnails, packaging, anything with words: Nano Banana Pro. Text accuracy is its main strength.
• Print or large displays: Nano Banana Pro at 4K.
• High volume social images or prompt exploration: Nano Banana 2 Lite. It is cheap and fast, and accepts up to 10 reference images.
• Quick, natural-looking photo edits: Nano Banana.
• Transparent backgrounds for logos or stickers: GPT Image 1.5, which supports an optional transparent background.
• Lowest cost per image with high resolution options: Flux 3 Klein, covered in our FLUX 3 Klein guide.

If you are not sure, try the same prompt across several models in the image playground before writing code. Ten minutes of side-by-side comparison usually settles it.

How to call Nano Banana Pro through the API

Unicorn API runs image models as jobs. You create a job with POST /v1/jobs, then poll GET /v1/jobs/{id} until its status is succeeded, and read the image URL from outputs[0].url. First, create an API key and export it as UNICORN_API_KEY.

Quick test with curl

curl -X POST https://api.unicornapi.net/v1/jobs \
  -H "Authorization: Bearer $UNICORN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "nano-banana-pro",
    "input": {
      "prompt": "A vintage coffee shop poster that reads \"OPEN LATE\" in bold cream letters, warm evening light",
      "resolution": "2K",
      "aspect_ratio": "4:3"
    }
  }'

# Then poll with the id from the response:
curl https://api.unicornapi.net/v1/jobs/JOB_ID \
  -H "Authorization: Bearer $UNICORN_API_KEY"

A complete Python script

This script reads the model id from an environment variable, which is the detail that matters for the next section.

import os
import time
import requests

API = "https://api.unicornapi.net"
HEADERS = {"Authorization": f"Bearer {os.environ['UNICORN_API_KEY']}"}
MODEL = os.environ.get("IMAGE_MODEL", "nano-banana-pro")


def generate(prompt, max_wait=300, **options):
    body = {"model": MODEL, "input": {"prompt": prompt, **options}}
    r = requests.post(f"{API}/v1/jobs", json=body, headers=HEADERS, timeout=30)
    r.raise_for_status()
    job = r.json()

    deadline = time.time() + max_wait
    while job["status"] not in ("succeeded", "failed"):
        if time.time() > deadline:
            raise TimeoutError(f"Job {job['id']} still running after {max_wait}s")
        time.sleep(3)
        r = requests.get(f"{API}/v1/jobs/{job['id']}", headers=HEADERS, timeout=30)
        r.raise_for_status()
        job = r.json()

    if job["status"] != "succeeded":
        raise RuntimeError(f"Job {job['id']} ended with status {job['status']}")
    return job["outputs"][0]["url"]


if __name__ == "__main__":
    url = generate(
        "A red fox resting in a snowy forest at dawn, soft light, highly detailed",
        resolution="4K",
        aspect_ratio="16:9",
    )
    print(url)

That 4K image costs 24 credits. Set IMAGE_MODEL=nano-banana-2-lite and drop the resolution argument, and the same script produces 4-credit drafts. If you work inside Cursor or Claude Code, you can also call these models from your editor through our MCP integration.

How to be ready for nano banana 3 on day one

You cannot use nano banana 3 yet, but you can make adopting it cheap. Most of the cost of switching models is code that assumes one model’s quirks. Avoid that now.

Keep the model id in configuration

Never hardcode nano-banana-pro deep inside your app. Read it from an environment variable or a config file, as the script above does. When a new model appears in the catalog, you test it by changing one value.

Keep per-model options separate

Input fields differ between models. Nano Banana Pro uses resolution, the original Nano Banana uses image_size, Nano Banana 2 Lite has neither, and Flux 3 Klein uses num_images where the Nano Banana models use n. Store options in a small per-model map rather than passing the same dictionary everywhere. A future nano banana 3 model may introduce its own fields too.

Build your test set now

Save the 10 to 20 prompts that matter most to you, along with your current best outputs. On release day, rerun them with the new model and compare. Without a saved baseline, you will judge the new model on memory and hype.

Mistakes people make

• Paying for “Nano Banana 3” from a reseller. If Google has not released it, you are paying for a relabelled existing model.
• Running every draft on the Pro tier. As the worked example shows, exploring on a 4-credit model and finishing on Pro cuts cost sharply.
• Choosing 1K on Nano Banana Pro to save money. 1K and 2K cost the same 18 credits, so ask for 2K unless you have a reason not to.
• Sending fields a model does not accept. Check the model page for its input fields before reusing a request body.
• Polling without a timeout. Always stop waiting after a sensible limit and log the job id, so a stuck request never hangs your pipeline.
• Switching production on launch day. New models often behave differently on the prompts you rely on. Compare first, then switch.

The short version: nano banana 3 is a name to watch, not a model to use yet. Nano Banana Pro is the best Google image model available today, Nano Banana 2 Lite covers cheap volume, and code that treats the model id as a setting will be ready for whatever Google ships next.