GPT Image 2.5 vs Nano Banana 2.1: where the price lines cross

On Sume, Nano Banana 2.1 is cheaper than GPT Image 2.5 in only two places: GPT at xhigh ($0.117125 per square image) and GPT at max ($0.2635) both cost more than Nano Banana 2.1 at 1K ($0.10). GPT Image 2.5 at low, medium and high costs less than every Nano Banana 2.1 tier up to 1K, so the useful question is which quality you need, not which model name is cheaper.

Every figure below is the Sume billable price: provider list times 1.25, rounded up to a micro-dollar. The numbers come from the pricing estimators in the Sume repository, for a 1:1 request with a short prompt and no reference images.

What does each tier cost per image?

GPT Image 2.5 has five quality tiers (low, medium, high, xhigh, max) plus auto, which reserves the max price. If you omit quality, the default is high. Nano Banana 2.1 is priced by resolution instead of quality.

Where exactly do the lines cross at 1K?

Nano Banana 2.1 at 1K is the closest match to a 1024 by 1024 GPT render. Against it, GPT high is $0.034125 cheaper per image, GPT xhigh is $0.017125 dearer, and GPT max is $0.1635 dearer.

Even the smallest Nano Banana 2.1 tier, 0.5K at $0.075, costs more than GPT high ($0.065875). If you were choosing Nano Banana 2.1 only because it looked like the budget option, check whether GPT Image 2.5 medium or high already meets the bar.

Does the answer change at 2K-class sizes?

Yes, a little. With the 2k size preset (2560 by 1440), GPT Image 2.5 costs more per tier than at 1024 square, but it still undercuts Nano Banana 2.1 at 2K ($0.15) up to xhigh.

Only max flips the result. At $0.276375 it costs more than Nano Banana 2.1 at 4K ($0.20).

How do I pick for a budget?

Decide the tier first, then divide. For 500 images, GPT low costs about $3.69, GPT medium $8.25, GPT high $32.94, Nano Banana 2.1 1K $50.00 and Nano Banana 2.1 4K $100.00. Sume also bills each request separately, so a per-request amount can round up to a whole cent in your wallet; treat the table as the exact micro-dollar value and keep a small margin when you plan a large batch.

Quality and resolution are not the same lever, so compare outputs on your own prompts before you commit a batch. The table only tells you what each setting costs.