Nano Banana 2 Lite Review: Testing Google's "Fastest, Most Cost-Efficient" Claim
Ship a Full-Stack App with One Prompt
Copy this prompt into your AI coding agent, or open it in one below.
Give this<br>to your AI
Create a to-do list app using Puter.js
Try in
Claude Code<br>Codex<br>Claude<br>OpenAI<br>Cursor<br>Lovable
Coding manually? see the guide
On this page<br>Testing the Speed ClaimWhat We FoundPricingConclusionTry It YourselfRelated
Blog
Nano Banana 2 Lite Review: Testing Google's "Fastest, Most Cost-Efficient" Claim
Reynaldi Chernando
July 14, 2026
On this pageTesting the Speed ClaimWhat We FoundPricingConclusionTry It YourselfRelated
We recently added Nano Banana 2 Lite to Puter.js. It's the marketing name for Gemini 3.1 Flash Lite Image, and Google recommends it as the replacement for the original Nano Banana (Gemini 2.5 Flash Image).
Google's pitch in the announcement is that this is their "fastest, most cost-efficient Gemini Image model", generating a 1K image in about 4 seconds. We tested that claim.
Testing the Speed Claim
Since the claim is about Google's own image family, we compared Nano Banana 2 Lite against the three other Nano Banana models on Puter.js. Those are Nano Banana 2, Nano Banana Pro, and the original Nano Banana it replaces.
We used the same prompt for all four and timed each one end to end from the browser, averaging three runs each. These are the times a real user waits for, network and queueing included, not raw GPU time.
The prompt is built to exercise the three qualities Google says the Lite model keeps, which are prompt adherence, character consistency, and legible in-image text rendering. It asks for the same character three times in one image, plus rendered text:
const prompt = "A children's book page showing the same red fox in a blue scarf three times: reading under a tree, napping on a rock, and chasing a butterfly, with the title 'The Quiet Hour' hand-lettered at the top";
const models = [<br>{ label: "Nano Banana 2 Lite", opts: { model: "gemini-3.1-flash-lite-image" } },<br>{ label: "Nano Banana", opts: { model: "gemini-2.5-flash-image" } },<br>{ label: "Nano Banana 2", opts: { model: "gemini-3.1-flash-image-preview" } },<br>{ label: "Nano Banana Pro", opts: { model: "gemini-3-pro-image-preview" } },<br>];
async function runModel(opts, runs = 3) {<br>const times = [];
Show 28 more lines...
const prompt = "A children's book page showing the same red fox in a blue scarf three times: reading under a tree, napping on a rock, and chasing a butterfly, with the title 'The Quiet Hour' hand-lettered at the top";
const models = [<br>{ label: "Nano Banana 2 Lite", opts: { model: "gemini-3.1-flash-lite-image" } },<br>{ label: "Nano Banana", opts: { model: "gemini-2.5-flash-image" } },<br>{ label: "Nano Banana 2", opts: { model: "gemini-3.1-flash-image-preview" } },<br>{ label: "Nano Banana Pro", opts: { model: "gemini-3-pro-image-preview" } },<br>];
async function runModel(opts, runs = 3) {<br>const times = [];<br>let image;<br>for (let i = 0; i element<br>times.push(performance.now() - start);<br>const mean = times.reduce((a, b) => a + b, 0) / times.length;<br>return { seconds: mean / 1000, image };
(async () => {<br>// run every model at once, then render in order once they all finish<br>const results = await Promise.all(<br>models.map(async ({ label, opts }) => ({ label, ...(await runModel(opts)) }))<br>);
for (const { label, seconds, image } of results) {<br>const card = document.createElement("figure");<br>const caption = document.createElement("figcaption");<br>caption.textContent = `${label}: ${seconds.toFixed(2)}s`;<br>image.style.maxWidth = "320px";<br>card.append(caption, image);<br>document.body.appendChild(card);<br>})();
Collapse code
You can run it yourself in the Puter playground, or drop it into a page on your own site.
What We Found
Average seconds per image over three runs, lower is faster:
Model<br>Average time per image
Nano Banana 2 Lite<br>3.23 s
Nano Banana<br>11.13 s
Nano Banana 2<br>7.73 s
Nano Banana Pro<br>17.02 s
Nano Banana 2 Lite came in at 3.23 seconds per image, about 2.4x faster than Nano Banana 2, the next fastest model in the set. Against the original Nano Banana it replaces, it was about 3.4x faster. Nano Banana Pro, the highest-quality tier, was the slowest at 17.02 seconds, about 5.3x slower than Lite.
Google's announcement says about 4 seconds per image. In our runs it averaged 3.23 seconds, under the stated figure, so the claim holds.
Image quality is subjective, so generate a few and judge for yourself. Judged on Google's own criteria for the model (prompt adherence, character consistency, and in-image text rendering), the output hit all three in our runs. The scene matched the prompt, the fox stayed consistent across the three poses, and the title text rendered legibly.
Here is each model on the same prompt:
Nano Banana 2 Lite
Nano Banana
Nano Banana 2
Nano Banana Pro
Pricing
Here is the per-image price for each model on Puter.js:
Model<br>Price per image
Nano Banana 2 Lite<br>$0.0336
Nano...