Google Deep MindOctober 6, 2026

Nano Banana 2.1

Google Nano Banana 2.1

Google’s image generation and editing model with up to 4K output; Standard 1K image output costs $0.0336 per image.

Text + Image + Video inputText + Image outputClosed / API only

At a glance

Context window
128K tokens
Max output
32K tokens
Text input / output · per 1M tokens
$1.50 / $7.50
Knowledge cutoff
—

API pricing

Text rates in USD per million tokens; other billing units shown below · effective 2026-10-06

Input
$1.50
Output
$7.50

Blended price4:1 input:output · same basis as the price ranking

$2.70/ 1M tokens

Service tiers

USD per million text tokens for additional published service tiers.

TierInputOutputCache read
Batch$0.75$3.75—
  • Source: https://ai.google.dev/gemini-api/docs/pricing Source: https://ai.google.dev/gemini-api/docs/pricing; includes thinking tokens.

Image pricing

  • image input$1.50 / 1M tokens

    Source: https://ai.google.dev/gemini-api/docs/pricing

  • image output$30.00 / 1M tokens

    Source: https://ai.google.dev/gemini-api/docs/pricing; image output only: 1K=1120, 2K=1680, 4K=2520 tokens; input/text/search billed separately.

  • image input$0.75 / 1M tokens

    batch · Source: https://ai.google.dev/gemini-api/docs/pricing

  • image output$15.00 / 1M tokens

    batch · Source: https://ai.google.dev/gemini-api/docs/pricing; image output only: 1K=1120, 2K=1680, 4K=2520 tokens; input/text/search billed separately.

Estimate running cost for Nano Banana 2.1

Specifications

Provider
Google Deep Mind
Model type
Vision model
Released
October 6, 2026
Architecture
Undisclosed
Context window
128K tokens
Max output
32K tokens
Input
Text, Image, Video
Output
Text, Image
Weights
Closed / API only
Commercial use
Proprietary

Capabilities

Reasoning

Model analysis

Nano Banana 2.1 is Google DeepMind’s image generation and editing model, released on October 6, 2026 as gemini-nano-banana-2.1. Its model card identifies Gemini 3.6 Flash as the base. Image-output tokens cost half as much as Nano Banana 2, while input tokens cost three times as much: reference-heavy editing does not automatically become cheaper.

Lower image-output costs do not cover the whole request

Gemini API pricing lists Standard input at $1.50 per million text, image or video tokens, text/thinking output at $7.50, and image output at $30. Image-only output costs are $0.0336 at 1K, $0.0504 at 2K and $0.0756 at 4K; Batch halves these rates. Nano Banana 2 charges $60 per million image-output tokens and $0.50 for input. Input, text output and search are additional. Search grounding includes 5,000 monthly requests shared across Gemini 3.x, then costs $14 per 1,000 searches.

The API limits differ from the model card

The API documentation specifies 131,072 input and 32,768 output tokens, with text, image, video and PDF inputs and text/image outputs. The model card instead describes up to 1M context and 64K text output. The official sources do not explain this difference; this page uses the API limits. Output supports 1K, 2K and 4K, with 1K as default, up to 14 reference images, and minimal, medium (default) and high thinking levels.

Arena results show progress but remain preliminary

On October 6, Arena text-to-image lists 2.1 fifth at 1328±9 from 5,312 votes, versus Nano Banana 2 at 1261±4 from 59,779. Single-image editing lists 2.1 sixth at 1428±6 from 12,985 votes, versus 1387±3 from 229,782 for its predecessor. Both 2.1 results are Preliminary. These preference scores are not accuracy rates or comparable to Google’s internal evaluation scale.

Small text and spatial editing remain limitations

Google reports improvements in visual quality, text rendering and consistency. Its model card still acknowledges blurry small text, weak long-paragraph rendering, imperfect character consistency, incomplete masked-edit instructions and left/right confusion. Consider 2.1 for iterative editing and high-resolution output; compare GPT Image 2.5 when higher Arena preference matters, although the leaderboard does not establish a cost-matched comparison. Parameter counts and an exact supported-language count remain undisclosed; weights are not downloadable.

Last reviewed 2026-10-06

Community discussion

Share the API, setup, and tradeoffs you actually tested. Rate the model once; comment as often as you like.

Used this model? Sign in with Google to rate it and share what worked, what did not, and which API or setup you tested.

No community comments yet.