Gemini Nano Banana 2.1 Goes GA: What a 4K Image Costs

Google made Gemini Nano Banana 2.1 generally available on October 6, 2026, and the number that matters for your budget is on the pricing page: a 4K image costs $0.113 on the standard tier and $0.0567 on Batch. There is also no free tier, so every test image gets billed.

What shipped with Gemini Nano Banana 2.1

The model ID is gemini-nano-banana-2.1. According to the Gemini API changelog, it updates Nano Banana 2 (gemini-3.1-flash-image) with better visual quality, prompt adherence and text rendering, at Flash-level speed and cost.

The older model is now deprecated, and Google tells developers to migrate. No shutdown date has been announced yet.

Nano Banana 2.1 pricing per image

Image output is billed per token: $30 per million tokens on the standard tier and $15 per million on Batch. Each resolution consumes a fixed number of tokens, so you get a flat price per image. These figures come from the Gemini API pricing page.

| Resolution | Tokens per image | Standard | Batch | | --- | --- | --- | --- | | 1K (1024x1024) | 1,120 | $0.0336 | $0.0168 | | 2K (2048x2048) | 1,680 | $0.0504 | $0.0252 | | 4K (4096x4096) | 3,780 | $0.113 | $0.0567 |

These prices cover image output only. Your prompt is billed as input ($1.50 per million tokens standard, $0.75 on Batch), and text or thinking output is $7.50 per million standard and $3.75 on Batch.

If you see a different 4K number on a reseller or tracker site, check it against the pricing page. 3,780 tokens at $30 per million is $0.113.

Is Nano Banana 2.1 cheaper than gemini-3.1-flash-image?

For image output, yes. The same pricing page lists the old model at $0.067 (1K), $0.101 (2K) and $0.151 (4K) on the standard tier. That makes 1K and 2K images about half the price, and 4K about a quarter cheaper.

4K saves less because the new model uses 3,780 tokens for a 4K image, against 2,520 on the old one. Halving the rate does not halve the bill there.

The catch is everything around the image. Input went from $0.50 to $1.50 per million tokens, and text and thinking output from $3 to $7.50. A workflow that sends long prompts or many reference images per result will save less than the table suggests.

What image generation does to a monthly budget

Take 10,000 images a month, counting image output only:

• 1K: $336 standard, $168 on Batch.
• 2K: $504 standard, $252 on Batch.
• 4K: about $1,130 standard, about $567 on Batch.

Resolution is the biggest lever you control. A 4K image costs more than three times a 1K image, so ask who actually needs 4K before it becomes the default.

No free tier: plan for it

The pricing page marks the free tier as not available for both input and output on the standard and Batch tables. Prototyping needs a billing-enabled project. Set a spend alert before someone puts an image generator inside a loop.

Monday morning: migrate and cut the bill

1. Search your code and config for gemini-3.1-flash-image. No shutdown date today does not mean none is coming.
2. Move work that can wait, like catalog images, campaign variants and overnight jobs, to Batch for the 50% discount. The Gemini API docs cover how to submit batch jobs.
3. Make 1K or 2K the default and treat 4K as an explicit request.
4. Re-run your prompt set against the new model before switching production traffic. Outputs will differ, and text rendering is one of the things Google says improved.

The image generation docs show the call through the Interactions API. This is the Python example, with the resolution set in response_format:

# pip install google-genai
# The client reads your key from the GEMINI_API_KEY environment variable
from google import genai
import base64

client = genai.Client()

interaction = client.interactions.create(
    model='gemini-nano-banana-2.1',
    input='<your prompt here>',
    response_format={
        'type': 'image',
        'aspect_ratio': '16:9',
        'image_size': '2K',
    },
)

with open('output.png', 'wb') as f:
    f.write(base64.b64decode(interaction.output_image.data))

The image_size value is the cost dial, so route it from config instead of hardcoding it.

Gotchas

• image_size needs an uppercase K. 2K works and 2k is rejected.
• The pricing page lists 1K, 2K and 4K for the new model, with no 0.5K tier like the old model had.
• Every generated image carries a SynthID watermark, which matters if your legal team has views on provenance.
• Grounding with Google Search has its own per-request pricing on the same page.

Image generation now has a clear per-image price, so you can put it in a budget line before it shows up as a surprise on the invoice.