Google released Nano Banana 2.1, an image generation and editing model, on 6 October 2026. Google's model card for the model is dated 6 October 2026, and its Google Cloud documentation lists a release date of October 6, 2026 and a launch stage of GA (generally available). Everything below comes from Google's own documents and two outlets' reports of Google's pricing page; none of it is independent testing.
What Google says it is
According to the model card, Nano Banana 2.1 is a member of the Gemini 3 series and is based on Gemini 3.6 Flash. It takes text and images as input and produces image and text output. Google's Cloud documentation describes it as optimized for multimodal image generation and editing, and says it "offers a balance of price and performance." The model ID there is gemini-nano-banana-2.1.
The model card lists seven distribution channels: Gemini App, Google AI Studio, Gemini API, Google Search AI Mode, Google Ads, Google Flow and Google Stitch. The Google Cloud documentation covers the model on the Gemini Enterprise Agent Platform.
The documentation lists three output resolutions: 1K, 2K and 4K. It also lists these features as supported:
- thinking
- multi-turn image editing
- grounding with Google Search and image search
- Content Credentials (C2PA)
- virtual try-on
- person generation
The following are listed as not supported:
- structured output
- function calling
- code execution
Two documents, two sets of limits
Google's two documents disagree on the model's size limits.
- Context window: the model card says "up to 1M"; the Google Cloud documentation lists 131,072.
- Output: the model card gives a 64K token output for text; the Cloud documentation lists a maximum of 32,768 output tokens.
The model card is for the model and the Cloud page is for its listing on Google's enterprise platform, so the two may describe different things. Neither document explains the difference. Developers building on it should check which limit applies on the platform they use.
Pricing
Neither the model card nor the Cloud documentation page lists prices. Unite.AI reports that Google's Gemini API pricing page, updated 7 October 2026, lists paid-tier image output for Nano Banana 2.1 at $30.00 per million tokens. The same page, as Unite.AI reports it, lists the predecessor, Gemini 3.1 Flash Image (Nano Banana 2), at $60.00 per million image tokens. Unite.AI quotes the page as describing the new model as "an update to Nano Banana 2 (Gemini 3.1 Flash Image)".
The Cloud documentation gives the tokens an output image consumes:
| Resolution | Output image tokens |
|---|---|
| 1K | 1,120 |
| 2K | 1,680 |
| 4K | 3,780 |
Unite.AI reports the pricing page's per-image equivalents as follows:
| Resolution | Nano Banana 2.1 | Nano Banana 2 |
|---|---|---|
| 1K | $0.0336 | $0.067 |
| 2K | $0.0504 | $0.101 |
| 4K | $0.113 (as reported; see below) | $0.151 |
The per-token rate for image output is halved, and the 1K and 2K per-image figures halve with it. Unite.AI's 4K figure of $0.113 does not: it matches the 3,780 tokens Google Cloud's documentation counts for a 4K image, while the $0.151 for Nano Banana 2 matches 2,520 tokens, the 4K count CellCog reports from the Gemini API pricing page. CellCog's report of Google's pricing page, as of 6 October 2026, gives $0.0756 at 4K: 2,520 tokens at $30 per million, half of Nano Banana 2's $0.151. Unite.AI also reports that the page lists no free tier for Nano Banana 2.1.
Google's own benchmark results
The model card's results are Google's own, from side-by-side human evaluations producing Elo scores and a single-sided AutoRater for factuality. Google describes them as results as of October 2026. For text-to-image Overall Preference, the card reports:
- Nano Banana 2.1 (Thinking): 1050 ±14
- Nano Banana 2.1 (No Thinking): 1015 ±13
- Gemini 3.1 Flash Image, "Nano Banana 2" (Thinking): 990 ±7
- Gemini 3 Pro Image, "Nano Banana Pro": 935 ±8
On Infographic Factuality, the card lists 0.521 for Nano Banana 2.1 (Thinking), against 0.179 for Nano Banana 2 and 0.265 for Nano Banana Pro. In the editing table, the Thinking configuration is listed above both earlier models on every row, including Multi-Character Consistency (1106 ±14) and Mask/Ink-Based Editing (1049 ±15). The Unite.AI report on the card does not cite any independent testing of these figures.
Limitations Google lists
The model card says Nano Banana 2.1 may show general limitations of foundation models, such as hallucinations, and that there may be occasional slowness or timeout issues. It lists these areas for improvement:
- small text that is often blurry in the 1K model, long paragraphs and page-length text
- imperfect character consistency between input and generated images
- partial instruction following and ink persistence in masked or doodle-based editing
- rare cases of persistent subject pose during editing
- occasional confusion over spatial position, such as left and right
- limited advanced capabilities in world knowledge, 3D reasoning and factuality
The card says the knowledge cutoff for Gemini 3.6 Flash is March 2026, and that in some domains knowledge may be limited to January 2025.
On safety, the card says specialist red teams outside the model development team found no egregious concerns when compared with Gemini 3.1 Pro. It also says the model satisfied required child-safety launch thresholds. For frontier safety, Google says it relied on its evaluations of Gemini 3.1 Pro and Gemini 3.7 Flash, neither of which reached any Tracked or Critical Capability Levels. It says it is confident Nano Banana 2.1 is also unlikely to reach any.
Sources
CellCog (2026) 'Nano Banana 2.1: Price, Arena Rank and What Changed', CellCog, 6 October 2026, available at: https://cellcog.ai/blog/nano-banana-2-1/ (Accessed: 14 October 2026).
Google Cloud (2026) 'Gemini Nano Banana 2.1 | Gemini Enterprise Agent Platform', Google Cloud documentation, last updated 13 October 2026, available at: https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/gemini/nano-banana-2-1 (Accessed: 14 October 2026).
Google DeepMind (2026) 'Nano Banana 2.1 — Model Card', Google DeepMind, 6 October 2026, available at: https://deepmind.google/models/model-cards/nano-banana-2-1/ (Accessed: 14 October 2026).
Reeve, J. (2026) 'Nano Banana 2.1 Halves the Per-Token Image Price of Its Predecessor', Unite.AI, 6 October 2026, available at: https://www.unite.ai/nano-banana-2-1-debuts-at-half-the-image-cost-of-its-predecessor/ (Accessed: 14 October 2026).