5,000 images at 2K: Google batch $126, Sume $750 on Nano Banana 2.1

The short answer

For 5,000 Nano Banana 2.1 images at 2K, Google's pricing page gives $126.00 at the batch rate (5,000 x $0.0252) and $252.00 at the standard rate (5,000 x $0.0504). Sume bills $750.00 for the same count (5,000 x $0.15). GPT Image 2.5 medium at 2K on Sume is $278.13.

That is a factor of 5.95 against Google batch and 2.98 against Google standard. What Sume offers for that premium is one wallet, one API shape for many models, and results that come back as Sume-hosted URLs.

The arithmetic

Sume does not document a half-price batch tier for the Image API. The docs describe n up to 10 per call and an async mode, and the price is cost_usd times n. So the comparison to Google's batch column is a price comparison, not a feature match.

The GPT row is the one that undercuts Google standard by a small amount: $278.13 against $252.00 is 10 percent more, not three times more.

How to run it

At this volume use n set to 10 so the job is 500 calls, and use async or webhook mode so no client holds a connection open for the 30-second budget.

If the job is bulk and time-insensitive and your account can use Google directly, Google's batch rate is the lower price. Use Sume when you need the other models in the same run.

Retries and what to watch

At 5,000 images the bill moves in blocks of 1,000. Each 1,000 images at 2K is $126.00 / 5 = $25.20 at Google batch, $50.40 at Google standard, $55.63 at GPT Image 2.5 medium on Sume and $150.00 at Nano Banana 2.1 on Sume.

Use those block prices to decide where a mixed job should run.

• Bulk drafts where any model will do: the lowest block price wins.
• Finals where one model is required: price that model alone, because the block price of the cheaper row does not apply.
• Anything that needs Sume-only features, such as mask_url on GPT Image 2.5: run it on Sume and keep the rest elsewhere.

What the docs say to check

Check the status code, not the body shape. POST /v1/images waits up to 30 seconds and returns the images with 200. Slow settings (4K, high quality, a large n) are the most likely to come back as 202 with a job envelope, and you then read the images from the job result endpoint described in Jobs and results.

The endpoint pricing lines are what Sume charges your wallet and already include the Sume margin, so a call costs cost_usd times n. Sume bills image generation all or nothing: a generation that fails or is cancelled is not charged.

Re-run the arithmetic with your own mix of tiers. Most bulk jobs use 1K, where the factors are the same as at 2K.

The numbers above are list prices read on 2026-10-08, not measurements. Sume figures are the catalog prices that already include its margin, Google figures are from its own pricing page, and quality or speed differences are not covered. Re-read both pages before you commit a large budget, because either side can change its rates.