Applied artificial intelligence, neural systems, and smart agents.
Fusing generative AI, multi-agent frameworks, and high-throughput data processing to automate business workflows.
Nano Banana 2
Nano Banana 2 is the gemini-3.1-flash-image unit: a 4K resolution ceiling, ten-image object reference capacity, and a $0.067 rate per 1K image. It holds rank 7 on the Arena text-to-image benchmark while Nano Banana Pro holds rank 12, meaning the lower-cost Google unit occupies the higher-ranked position.
- up to 4K
- ten object references
- $0.067 per 1K image
Last checked: This is a reference page. It is re-checked against the vendor sources and updated when a new version ships. Change log
One real gap, one statistical tie
On the text-to-image benchmark, Nano Banana 2 logs a score of 1263.8 against Nano Banana Pro at 1246.0. The gap runs about eighteen score points, and the two confidence intervals do not overlap: this unit interval begins at 1258.7 and Pro interval ends at 1249.4. This is therefore a genuine performance gap, not statistical noise.
On the editing benchmark the pattern reverses and must be recorded precisely: Pro scores 1389.1 against this unit 1385.1, putting Pro four points ahead. But the confidence intervals overlap in full (1381.0 to 1389.1 against 1386.1 to 1392.0), so on the editing axis the two units are statistically equivalent, with no citable advantage either way.
Set these two results against the pricing specification. A 1K image costs $0.067 on this unit and $0.134 on Pro — exactly half, for output that wins on one axis and ties on the other.
Operating constraint: the half-cost figure holds only at one resolution tier
This unit provisions four size tiers against Pro two. Lined up: at 0.5K the cost is $0.045, a tier Pro does not offer at all. At 1K the ratio is $0.067 to $0.134, exactly half. At 2K it is $0.101 against the same $0.134, since Pro applies one flat rate across both 1K and 2K. At 4K it is $0.151 against $0.24.
So the cost saving compresses as the output tier scales up: half at 1K, roughly three-quarters at 2K, roughly sixty percent at 4K. Our price column is built on the 1K tier because it is the only tier both units publish a rate for. For a workload running consistently at 2K, the saving shown in the table exceeds the saving realised on the invoice.
Ten-image reference capacity, and the two adjacent figures
Google publishes three separate ceilings for this unit: up to ten object images, up to four character images, and up to three style images. Our reference column is built on the object figure, and on that figure this unit tops the table. But for a workload requiring several consistent people in one scene, the governing figure is four, and Pro provisions five on that same axis — the one published, citable advantage Pro holds.
Which configuration actually earned the rank
The Arena row identifier reads "gemini-3.1-flash-image (nano-banana-2) [web-search]". That suffix is not decorative: the figure recorded in our table belongs to the search-grounded configuration. Invoking the same unit without grounding does not reproduce the measured configuration. This is the pattern that recurs throughout this reference: a leaderboard identifier names a serving configuration, not merely a model.
The Lite variant, frequently conflated with this unit
Google operates a third unit, gemini-3.1-flash-lite-image, listed on Arena as Nano Banana 2 Lite at rank 11. That unit supports only the 1K tier, accepts no character or style references, and Google states plainly that it is not optimised for multi-reference input or multi-turn sequential editing. In exchange it accepts up to fourteen object images, the highest figure of the three units. It is not a candidate in our table; for the lowest possible operating cost, evaluate it with these constraints understood.
From Iran
The Gemini API available-regions document was read today and Iran is not recorded on it. This finding concerns the API layer and makes no statement about the Gemini application. This platform does not sell accounts and does not propose a route around this restriction.
Specifications
Every number here comes from the vendor's own page, with that page linked beside it.
Using it from Iran
This column carries a date and says how it was checked, because most listicles guess it. Where we have only read a vendor policy page, the note below says exactly that.
- Reachable
- blocked
- Payment
- no working route
- Free tier
- no
- Measured on
How we checked: The Gemini API available-regions page lists the countries in which the interface is active, and Iran is not recorded on that list. This finding comes from reading a document, not from a network measurement.
Documented strengths
- Ranked above Google own flagship unit on text-to-image, by a margin the confidence intervals support
- Half the operating cost of Pro at the 1K tier, and lower cost across all three other tiers
- Ten-image object reference capacity, the highest figure among the table candidates
- Provisions a 0.5K tier that Pro does not, for workloads where a small output suffices
- Google describes its in-image text rendering as advanced, for infographics, menus and diagrams
Room for improvement and technical considerations
- It does not lead Pro on editing: four points behind, with intervals overlapping in full, indicating a statistical tie.
- The half-cost figure holds only at the 1K tier; by 4K the gap narrows to roughly sixty percent.
- Character reference capacity is four against five on Pro.
- Google publishes no pixel dimensions for any tier from 0.5K to 4K.
- No weights are published, and no local deployment route exists.
- Iran is not recorded on the Gemini API available-regions list.
- The figure recorded in our table belongs to the web-search row, so it is tied to a configuration, not the model alone.
Technical verdict
Between the two Google image units, this is the default configuration: lower operating cost, ahead on text-to-image, tied on editing. Provisioning Pro is justified when a fifth character reference image is a hard requirement, which is a specific workload most deployments do not carry. For a workload centred on editing, the top row of this table is not a Google unit at all.
Frequently raised questions, with documented answers
What is Nano Banana 2, technically
It is the commercial name for the Google image unit gemini-3.1-flash-image. Its resolution ceiling is 4K, its reference capacity reaches ten object images, and its deployment surface is the ordinary Gemini API.
What is the pricing structure of Nano Banana 2
On the standard tier: $0.045 at 0.5K, $0.067 at 1K, $0.101 at 2K, and $0.151 at 4K. The batch-processing route runs at roughly half of these rates.
How does Nano Banana 2 perform against Nano Banana Pro
On the text-to-image benchmark, 2 leads and the gap sits outside the confidence intervals. On the editing benchmark, the two units are statistically equivalent. At the 1K tier, the rate for 2 is half the rate for Pro.
What is the architectural difference between Nano Banana 2 and the Lite variant
Lite supports only the 1K tier and accepts no character or style references, but provisions capacity for up to fourteen object images. Google states explicitly that it is not optimised for multi-turn editing.
Is an access route available from Iran
Iran is not recorded on the Gemini API available-regions list, so no official route exists. This platform does not sell accounts and does not propose a route around this restriction.