Nano Banana 2.1 vs GPT Image 2.5 on Sume: 4.0x at 1K, 0.9x at 4K

On Sume's price list, Nano Banana 2.1 costs 4.04 times as much as GPT Image 2.5 at the 1K step, 2.70 times at 2K, and 0.90 times at 4K. At 4K the order flips: the Nano Banana 2.1 image is the cheaper of the two. The catch is that the GPT Image 2.5 rows are tiered by quality as well as size, so each step compares different quality settings. Read the ratios as a budgeting aid, not as a verdict on which model looks better.

People search for "nano banana 2.1 vs gpt image 2.5" right after the Nano Banana 2.1 launch on 2026-10-06. Most answers compare image quality. This page only does arithmetic on what Sume bills, because that is the part you can check against the public price list.

The three price pairs

Sume bills the provider list price times 1.25. Nano Banana 2.1 has one price per resolution tier. GPT Image 2.5 is tiered by quality and size, and the three points on the Sume list are low at 1K, medium at 2K, and high at 4K. Each ratio below divides the Nano Banana 2.1 price by the GPT Image 2.5 price at the same size step.

Why the ratio falls as size grows

Nano Banana 2.1 rises from $0.10 to $0.20 across the three steps, a factor of 2. GPT Image 2.5 rises from $0.02475 to $0.2225, a factor of 8.99. The list pairs a higher quality with each larger size, so that jump mixes both effects, and the Sume list does not show the off-diagonal points (for example low quality at 4K), so do not read them off this table.

A practical consequence: for a pipeline of small drafts, the 1K gap is the biggest, at roughly 4 to 1. For final 4K deliverables the two models are within about 10 percent of each other on price, so the choice at 4K has to rest on output rather than cost.

Check the price before you commit

The Image API returns the amount billed in usage.cost on every call, and GET /v1/images/models lists each model's endpoint pricing. A two-call comparison costs about 12.5 cents at the cheapest settings below.

The Image API accepts resolution for Nano Banana 2.1 and quality for GPT Image 2.5, as the Sume Image API docs describe. Sume rejects a parameter that a model does not list with 400 unsupported_parameter, so keep one request body per model.

curl -s https://api.sume.com/v1/images \
  -H "Authorization: Bearer $SUME_API_KEY" -H "Content-Type: application/json" \
  -d '{"model":"google/nano-banana-2.1","prompt":"a ceramic mug on oak","resolution":"1K"}'
curl -s https://api.sume.com/v1/images \
  -H "Authorization: Bearer $SUME_API_KEY" -H "Content-Type: application/json" \
  -d '{"model":"openai/gpt-image-2.5","prompt":"a ceramic mug on oak","quality":"low"}'
# Compare usage.cost in the two responses.

What this page does not claim

The page makes no claim about which model draws a better mug. The list has no price point for GPT Image 2.5 at high quality and 1K, or for low quality at 4K, so a like-for-like ratio at every size is not available from Sume's published rows. If you need one, send the exact request you plan to run and read usage.cost. Failed or cancelled generations are not billed, per the Image API billing rules.