Nano Banana 2.1 is the October 2026 update to Google's Nano Banana 2 (Gemini 3.1 Flash Image), adding 1:8 and 8:1 panoramic canvases and better multi-turn consistency for up to 4 characters. It suits developers and design teams that want one model for both generating and editing, and Google has already marked its predecessor deprecated.
What changed
- · Release Nano Banana 2.1 reaches general availability in the Gemini API (gemini-nano-banana-2.1) source
Nano Banana 2.1 (gemini-nano-banana-2.1) is Google DeepMind's image generation and editing model in the Gemini API, generally available since 6 October 2026. It outputs at 1K, 2K and 4K, accepts up to 14 reference images, and can ground a prompt in Google Search results.
Where it sits
- $3.00/M$ per 1M tokensBlended price (3:1)Lower is better#45 / 80peer median $1.86/Mvendor price, checked by HokAI
Priced around the middle of the 80 GA models with a published price (rank 45), and one of 77 whose vendor states it does not train on customer data.
Ranks are against GA models on HokAI that publish the same figure; ties share a rank.
Provider: Google DeepMind · Family: Nano Banana
More about Google DeepMind on HokAI
Input modalities: text, image, video · Output: image, text
About Nano Banana 2.1
Google DeepMind released its image generation and editing model Nano Banana 2.1 to general availability in the Gemini API on 6 October 2026 under the model ID gemini-nano-banana-2.1. Google frames it as a successor to the earlier Flash image model that keeps Flash-level speed and cost efficiency while improving visual quality, prompt adherence, multi-turn character consistency and text rendering. It also adds wide and panoramic canvases at 1:4, 4:1, 1:8 and 8:1 across 1K, 2K and 4K output. The same day, Google marked Nano Banana 2 as deprecated, with no shutdown date announced, and told developers to migrate.
The model takes text, images and video as input and returns images, with text and thinking output billed separately. Google's image generation guide says one request can mix up to 14 reference images, of which up to 10 objects can be kept at high fidelity and up to 4 characters can keep their appearance across edits. Audio input is not supported. Search grounding lets a prompt draw on Google Search web and image results, though Google says grounding does not use real-world photos of people from web search. Every output carries a SynthID watermark, Google's invisible marker for AI-generated images.
Inside Google's lineup, Nano Banana 2.1 is called the primary high-efficiency workhorse and is recommended for all new projects. The cheaper Nano Banana 2 Lite targets speed and volume but is not tuned for several reference inputs or multi-turn editing, while Nano Banana Pro, built on Gemini 3 Pro Image, is Google's pick for intricate compositions and brand-accurate assets. Outside Google, the closest model on HokAI is OpenAI's GPT Image 2, alongside FLUX.2, Ideogram and Adobe Firefly, and Midjourney suits artists who want a style-first workflow instead of an API. The guides to the best AI for creating images and videos, logo design and photo editing and restoration put the options side by side.
Developers reach the model through the Gemini API and Google AI Studio, and the Gemini page on HokAI covers the wider family. Google's pricing page lists a paid tier only, with no free tier for this model, and states that paid-tier content is not used to improve Google's products.
Who should look elsewhere: Google publishes no benchmark scores for this model on the pages HokAI read, so the quality gains over Nano Banana 2 rest on Google's own description, not on a leaderboard result. Teams that need exact brand assets, dense layouts or the most demanding localization are pointed by Google's own guidance toward Nano Banana Pro, and anyone who needs audio input or self-hosted weights cannot use this model at all. The context window and output token limits were not on the pages HokAI read, so they are left blank until Google publishes them.
Pricing
Google bills image output at $30 per 1M tokens on the Standard rate and $15 per 1M on Batch. A 1K image uses 1,120 output tokens ($0.0336), a 2K image 1,680 ($0.0504) and a 4K image 3,780 ($0.113). Text and thinking output is $7.50 per 1M, and input is $1.50 per 1M for text, image or video. Batch halves every rate. Search grounding is free for the first 5,000 requests a month, shared across Gemini 3 and newer models, then $14 per 1,000. The premium [Nano Banana Pro](/hub/models/gemini-3-pro-image) sits on a higher rate card. Source: Google's Gemini API pricing page.
What a real job costs
| Job | Input | Output | Total |
|---|---|---|---|
| Summarise a 20-page PDF | $0.045 | $0.0075 | $0.052 |
| Support reply | $0.0030 | $0.0022 | $0.0052 |
| One coding agent run | $0.300 | $0.150 | $0.450 |
Budgets: 20-page PDF = 30k in / 1k out · Support reply = 2k in / 300 out · Coding agent run = 200k in / 20k out. Computed from the vendor's per-token prices at render time; cached-input discounts are not applied.
Key Features
- Three output resolutions: Generates at 1K, 2K and 4K, where Google counts a 1K image as 1,120 output tokens and a 4K image as 3,780.
- Panoramic canvases: Adds tall poster and wide banner canvases, with the extremes at 1:8 and 8:1.
- Reference-image mixing: Mixes reference photos, objects and characters in one request so a product or person stays recognisable across edits.
- Character consistency: Holds the look of up to 4 characters across multi-turn edits.
- Search grounding: Pulls in Google Search web and image results while generating, billed per query after a monthly free allowance.
- SynthID watermark: Marks every output with Google's invisible SynthID signal, designed to identify AI-generated images.
Pros
- Costs about half as much per 1K image as Nano Banana 2 in Google's own table, so large batches get cheaper after migrating.
- One model handles both generation and conversational editing, so teams skip a separate editing endpoint.
- Google recommends it for all new projects, which makes it the safe default in the Nano Banana family.
- Batch mode halves the rate for work that does not need an instant reply.
Cons
- No benchmark scores or token limits appear on the Google pages HokAI read, so quality comparisons depend on your own tests.
- No audio input and no free tier in the API.
- Complex layouts and brand-accurate assets are routed by Google to Nano Banana Pro instead.
- Closed model: no weights and no self-hosting.
Frequently Asked Questions
How much does Nano Banana 2.1 cost in 2026?
Google's Gemini API charges $1.50 per 1M input tokens and $7.50 per 1M text and thinking output tokens, with image output at $30 per 1M tokens. That works out to $0.0336 for a 1K image, $0.0504 for 2K and $0.113 for 4K, and Batch halves every rate. Nano Banana 2 charges $0.067 per 1K image and Nano Banana Pro $0.134, so 2.1 is the cheapest of the three at 1K.
Can you use Nano Banana 2.1 without paying?
Not through the API: Google's pricing table lists the free tier as not available for this model, so every request is billed. Google AI Studio is linked from that page as a place to try it, but HokAI did not verify what usage limits apply there. Check the Gemini app's own limits separately if you only want casual use.
How does Nano Banana 2.1 compare to Nano Banana Pro?
Pick 2.1 for volume and speed: Google calls it the primary high-efficiency workhorse and recommends it for all new projects. Pick Pro for complex layouts, brand-accurate assets and the deepest world knowledge. The two also trade differently on references: Pro holds up to 5 characters and 6 high-fidelity objects, while 2.1 holds 4 characters and 10 objects.
Does Nano Banana 2.1 train on your data?
Google's pricing page marks the paid tier, the only tier for this model, as not used to improve its products. Several other Gemini models show the opposite for their free tier, so the paid-only status matters here. Outputs also carry an invisible SynthID watermark.
Who is Nano Banana 2.1 best for, and who should skip it?
It fits developers and design teams generating or editing many images, especially with recurring characters or unusual canvases. Skip it if you need audio input, self-hosted weights or published benchmark scores. Artists who want a style-first workflow are better served by Midjourney, and OpenAI-based stacks by GPT Image 2.
Top Alternatives
- Gemini 3 Pro Image: Pick Nano Banana 2.1 for volume and speed; pick Gemini 3 Pro Image for complex layouts, brand consistency and the deepest world knowledge.
- GPT Image 2: Pick Nano Banana 2.1 for Search grounding and 14-image reference mixing in the Gemini API; pick GPT Image 2 if your stack already runs on OpenAI.