Nano Banana 2.1 costs the same as Lite at 1K ($0.0336): what differs

Since October 6 there is no per-image saving in choosing Nano Banana 2 Lite over Nano Banana 2.1 at 1K: both list at $0.0336 per image on Google's pricing page. Lite is still cheaper on input tokens and on text and thinking output, and it is built for speed. It cannot go above 1K, has no search grounding, and is not tuned for many references.

Side by side

Both models bill image output at $30 per 1M tokens, and a 1K image is 1,120 tokens, so the image line matches. The other lines do not.

What the input gap is worth

A 500-token prompt adds $0.00075 on Nano Banana 2.1 and $0.000125 on Lite. That is under a tenth of a cent per image. Lite's edge only shows up when each call carries thousands of input tokens, such as long instructions or several references, and Google says Lite is not optimized for multiple reference inputs or multi-turn editing anyway.

How to choose

• Product shots, thumbnails and ads where text must be legible or facts must be current: Nano Banana 2.1.
• Square or 16:9 drafts at 1K in the thousands, where latency is the constraint: Lite.
• 2K or 4K output: Nano Banana 2.1, because Lite stops at 1K.

On Sume

Sume's Image API docs name Nano Banana 2.1 and Nano Banana Pro as its Google models, and I found no Lite entry there. If you need Lite, you call Google directly. Confirm what your key sees with GET /v1/images/models.