Google has rolled out Nano Banana 2 Lite, its fastest and cheapest AI image model yet. Priced at $0.034 per 1,000 images with roughly four-second generation, it targets high-volume commercial workflows. Surprisingly, it even out-scores the pricier Pro tier on one key benchmark.

Key Takeaways

• Nano Banana 2 Lite costs just $0.034 per 1,000 generated images
• It produces a 1K-resolution image in about four seconds
• It scored 1251 Elo, beating Nano Banana Pro’s 1245 mark
• The model replaces the original Nano Banana, now called legacy
• It launched alongside video model Gemini Omni Flash on June 30

What Nano Banana 2 Lite Actually Is

Google shipped Nano Banana 2 Lite on June 30, 2026, calling it the fastest and most cost-efficient model in the Nano Banana family. Its official API name is gemini-3.1-flash-lite-image. In plain terms, it’s the image-generation version of Google’s Gemini 3.1 Flash Lite model.

This is the third entry in the lineup. The original Nano Banana launched last summer on Gemini 3.1 Flash, followed by Nano Banana 2 in February with sharper realism. Google now describes the standard version as a “generalist workhorse,” while Lite is tuned purely for speed and scale.

The model is available immediately across three developer channels:

• Google AI Studio for interactive testing and prompt work
• The Gemini API for direct developer integration
• The Gemini Enterprise Agent Platform for business workflows

It’s also rolling out inside consumer products. According to the launch review, that includes AI Mode in Search, the Gemini app, NotebookLM, Google Photos, Stitch, Flow, and Google Ads.

One important note for existing users. Google now labels the original Nano Banana a “legacy model” and recommends Lite as its drop-in replacement. If you have an integration running on the old model (gemini-2.5-flash-image), plan to migrate soon.

The Pricing and Speed Play

The headline number is small on purpose. Nano Banana 2 Lite costs $0.034 per 1,000 images, a rate Google set to make large-scale image generation routine rather than a budget line item. That works out to well under four cents per thousand pictures.

Speed is the other half of the pitch. The model generates a 1K-resolution image in about four seconds, which suits teams that iterate fast or run automated pipelines. Google’s own developer blog frames the cost angle directly:

Cost-efficiency ($0.034 per 1K image): A cost-efficient choice for developers focused on drafting, ideating, managing operational budgets or low-bandwidth usage.

Here’s how Lite stacks up against the model it replaces:

So Lite is both cheaper and faster than the model it retires. VentureBeat notes the strategy is aggressive by design, undercutting the older, less capable model while binding usage to Google’s metered API pricing.

What does cheap really change? It shifts the whole workflow. Instead of carefully prompting one image at a time, teams can generate large option sets, filter automatically, and keep only the best. The smarter metric becomes cost per usable asset, not just cost per image.

The Benchmark Surprise

The “Lite” label suggests compromise. The benchmarks say otherwise. Nano Banana 2 Lite scored 1251 on the Text-to-Image Arena Elo, a standardized test based on blind human preference. That beats the legacy model’s 1151. More surprising, it edges out the premium Nano Banana Pro, which sits at 1245 on the same track.

The cheapest model in the family scoring higher than the flagship sounds like a mistake. It isn’t. Google credits targeted architecture tuning built for the 1K canvas: stronger world knowledge, better character consistency, and cleaner in-image text rendering. The model inherits the Gemini 3.1 generation’s instruction-following, so it wins at the common “one sentence, one image” task.

Editing scores hold up too:

• Single-image editing: 1308 Elo
• Multiple-image editing: 1294 Elo

A word of caution on all these figures. These scores are internally published by Google, not yet verified by independent evaluators like Artificial Analysis. Treat them as credible vendor claims worth testing against your own tasks.

There’s also a real ceiling. Lite supports only 1K output, while NB2 and Pro scale to 2K and 4K. So the Elo win comes with an asterisk. As one detailed breakdown put it, the single-image benchmark can’t capture Pro’s edge in 4K detail, complex multi-subject scenes, or realistic human faces. A single number shouldn’t decide your model.

Who Should Use It and When

The fit is clear once you match the model to the job. Nano Banana 2 Lite shines when you need lots of good-enough images fast, and when 1K resolution is acceptable. Google positions it as the immediate infrastructure workhorse for high-throughput commercial work.

Strong use cases include:

• Testing many image variations during creative work
• Generating batches of assets for marketing or product listings
• Ad variants, thumbnails, mockups, and UI concepts
• Automated pipelines where image output is one step among many

Pick a heavier model when the stakes are higher. If you need a 4K key visual for a client, complex multi-subject composition, or rock-solid realistic faces, Nano Banana 2 or Pro remains the safer choice. A sensible default: explore with Lite, finish with Pro.

There’s a strategic layer worth noting. The launch paired Lite with Gemini Omni Flash, a video model priced at $0.10 per second of output. Together they form a static-to-motion pipeline: draft an image with Lite, then animate it with Omni Flash.

The catch is lock-in. Unlike open-weight models you can run locally, Google keeps these tightly bound to its managed cloud and metered pricing. That removes hosting headaches but ties you to Google’s terms. For teams already in the Google ecosystem, that trade is easy. For everyone else, it’s the real question to weigh before committing.

For developers, the first test is simple. Run your existing prompts through the new model, then compare speed, cost, text rendering, and quality against your current setup before moving production traffic.

Digital Trendings is your trusted source for AI news and updates — stay tuned for more.