Nano Banana 2.1 vs Nano Banana Pro: API Price per Image and Which to Use

Pro costs 2.1 to 4 times as much per image

Google prices both models by image-output tokens, at $30 per million for 2.1 and $120 per million for Pro, and fixes the token count per resolution. Pro charges the same 1,120 tokens for 1K and 2K, so a 2K Pro image costs no more than a 1K one. The table gives Google's per-image figures for the paid tier of the Gemini API, image output only.

Read down the Pro column: even Pro's Batch price at 2K ($0.067) is above 2.1's Standard price ($0.0504). For an illustrative run of 1,000 images at 2K, image output costs $50.40 on 2.1 and $134 on Pro at Standard rates, or $25.20 and $67 through Batch; at 4K it is $113 against $240 (ToolColumn arithmetic). Reference images move the gap slightly the other way: Pro bills an input image at 560 tokens ($0.0011), while Agent Platform's footnotes put 2.1 at 1,120 tokens, about $0.0017 at $1.50 per million input tokens. Three references per request add about $5.04 per 1,000 images on 2.1 and $3.36 on Pro. Both models also bill thinking text as output ($7.50 per million on 2.1, $12 on Pro), and Google does not charge for the interim thought images.

Two price details are inconsistent on Google's own pages. The pricing page and Agent Platform bill a 2.1 4K image as 3,780 tokens ($0.113), but the size table in the image-generation guide lists 2,520 tokens, which would be about $0.076; the prices above follow the pricing page. For Pro, the pricing page lists a Flex rate equal to Batch and a Priority rate of $216 per million image-output tokens, yet Pro's Gemini API model page says Flex and Priority inference are not supported. On Agent Platform, Pro's model page lists Standard, Flex and Priority. For 2.1 the Gemini API offers neither Flex nor Priority, and Agent Platform sells it at Standard and Priority rates but says it does not support Flex.

Google's 2.1 model card puts 2.1 ahead of Pro on all ten rows

The only Google source that tests 2.1 against Pro is the Nano Banana 2.1 model card, which reports side-by-side human-preference Elo ratings as of October 2026, plus one automated factuality rating. The figures below are Google's; the last column is ToolColumn's subtraction.

With thinking, 2.1 leads Pro by 37 to 136 Elo points on the nine preference rows; Pro's best showing is single-character consistency. Without thinking, 2.1 still sits above Pro on every row, though product consistency (981 ±18 against 965 ±14) falls inside the card's own margins. Google does not say which API setting the "no thinking" column used, since its docs say thinking cannot be turned off for Gemini 3 image models; 2.1 offers minimal, medium (the default) and high levels.

Google's descriptions still lean toward Pro for hard jobs. Agent Platform calls Pro "the best model for complex and multi-turn image generation and editing", and the Gemini help center, which describes redoing Nano Banana 2 images, says Pro can add detail for text rendering and infographics. The 2.1 model card puts 2.1 ahead of Pro on both infographic rows. Test your own prompts before paying Pro's rate for those.

Pro keeps style references, a fifth character and temperature control

The capability differences below come from Google's image-generation guide and the two models' Gemini API and Agent Platform pages. Read the rows where only one model has the feature: those decide the choice regardless of price.

Pick Pro when a brief supplies style images, when a scene needs five recurring people, or when you want to vary results through temperature. Pick 2.1 to build images from video frames, to ground a picture in Google Image Search, or to choose a thinking level per request. Both cover aspect ratios from 1:8 to 8:1, multi-turn editing and C2PA Content Credentials on Agent Platform, and every Gemini API image carries a SynthID watermark. Agent Platform lists Pro's knowledge cutoff as January 2025.

Nano Banana 2 is deprecated, and 2.1 costs less at every size

For readers comparing 2.1 with the previous model: Google deprecated Nano Banana 2 (gemini-3.1-flash-image) on October 6, 2026, the day 2.1 became generally available, with no shutdown date announced, and its changelog says to migrate to gemini-nano-banana-2.1. The deprecated model costs $0.067, $0.101 and $0.151 per 1K, 2K and 4K image, so 2.1 is half its price at 1K and 2K and a quarter cheaper at 4K, although 2.1's input tokens cost $1.50 per million against $0.50. Reference limits (10 objects, 4 characters), video input and Image Search grounding carry over, and 2.1 adds the medium thinking level. The earlier model keeps two things 2.1 lacks: a 512px size at $0.045, and US and EU multi-regions on Agent Platform, where its retirement date reads "May 28, 2027 or later". Pro is not deprecated.

Where each model runs: API, Agent Platform, Gemini app and Flow

• Gemini API and Google AI Studio: gemini-nano-banana-2.1 (built on Gemini 3.6 Flash, per its model card) and gemini-3-pro-image are both stable. Google's pricing page lists the free tier as "Not available" for every Nano Banana model, so AI Studio image generation needs a paid key. The older gemini-3-pro-image-preview ID shut down on June 25, 2026.
• Gemini Enterprise Agent Platform (formerly Vertex AI): both are generally available in the global region at the Gemini API's Standard rates. 2.1 has Standard, Priority and Batch but no Flex, and no retirement date; Pro has Standard, Flex and Priority, with retirement "May 28, 2027 or later".
• Gemini app: gemini.google says Google AI Plus ($4.99 a month), Pro ($19.99) and Ultra ($99.99 or $199.99) get Nano Banana 2.1, free accounts get the earlier Nano Banana 2, and paid users can regenerate an image with Pro from the three-dot menu ("Redo with Pro"). Google's help center does not mention 2.1: it maps the app's Flash and Pro settings to the deprecated Nano Banana 2 and stops Pro redos once you hit the daily quota of Nano Banana 2 images. Google publishes no image counts per plan.
• Google Flow: Flow's help lists 2.1 as the standard model and Pro as the default on AI Ultra, and Google's plans page gives free accounts limited access to Pro in Flow. Google does not publish credits per image for either model.