Nano Banana 2.1 input is $1.50 per 1M tokens: when does it matter?

Nano Banana 2.1 input costs $1.50 per 1M tokens, three times the $0.50 that Nano Banana 2 charged. It still wins on total cost for a 1K image until a single request carries about 33,400 input tokens, because its output image costs $0.0336 against $0.067. Most prompt-only and few-reference calls are far below that.

The break-even arithmetic

The 1K total is the image price plus input tokens times the rate. Setting the two models equal gives (0.067 - 0.0336) / (1.50 - 0.50) per million, which is 33,400 tokens. At 2K the gap in image price is larger, so the crossing point is 50,600 tokens. These are arithmetic from the list prices, not measurements.

What to measure

Google's pricing page lists the input rate but not how many tokens one input image uses. Do not guess: send a typical edit with your real reference images and read the token counts your own response reports, then plug them into the formula. Text-only calls can skip this step.

Output has its own surprise

Text and thinking output is $7.50 per 1M on Nano Banana 2.1 against $3 on Nano Banana 2. If a request produces long text alongside the image, that line grows too. Google's image guide says the "thought images" the model makes while composing are visible in the backend but not charged.

On Sume

Sume bills image models per image, not per token: the Image API docs say usage.cost is the billed USD amount and token counts are 0 in v1. So the input-rate change does not reach a Sume invoice as a separate line. It is folded into the per-image price Sume lists.