Gemini image API price per image: Lite, Flash and Pro tiers

A Gemini image costs roughly $0.0336 on Gemini 3.1 Flash Lite Image at 1K, $0.045 to $0.151 on Gemini 3.1 Flash Image from 512px to 4K, and $0.134 to $0.24 on Gemini 3 Pro Image, according to Google's pricing page read on 2026-10-02. On Sume the figure that matters is the per-image cost_usd line returned by the image catalog, because the margin is already in it.

Google bills these models in tokens, so the per-image numbers are approximations the page itself prints. This post turns the token rates into a per-image ladder and shows where to read the equivalent number on Sume before you spend anything.

What does Google list per image for each Gemini image tier?

The page lists three image models. Each has an output token rate for images and a rounded per-image figure that depends on resolution.

Input tokens are billed on top: $0.25, $0.50 and $2.00 per million for Lite, Flash and Pro respectively. A prompt with several reference images therefore costs more than the output figure alone.

Why is the same tier cheaper at 512px and dearer at 4K?

Because the unit is a token. A larger image is more output tokens, so the per-image figure scales with resolution. Flash goes from $0.045 to $0.151, a factor of about 3.4 between the smallest and largest size Google prices. Pro barely moves between 1K and 2K ($0.134) and then rises to $0.24 at 4K.

The practical consequence is that the tier choice matters less than the size you ask for. A 4K Flash image ($0.151) costs more than a 1K Pro image ($0.134), so ask for the size your layout needs rather than defaulting to the largest.

How does Sume show the price of one image?

Sume's image surface returns pricing lines per model. Each line has a billable unit and a cost_usd, and the docs state that endpoint pricing lines are the amount charged to the wallet: the margin is already applied, so cost_usd times the number of images is what you pay. Image models shows a line shaped like { "billable": "output_image", "unit": "image", "cost_usd": 0.033 }.

Read the live lines for your workspace with the model list. Use the pricing block of the model you care about, not a figure from a blog post, including this one:

curl "https://api.sume.com/v1/images/models" \
  -H "Authorization: Bearer $SUME_API_KEY"

What would 100 images cost at list and at Sume's margin?

Sume's published rate card is provider list times a house margin, and the repo's standard margin ratio is 1.25. The arithmetic below is only an illustration of that multiplier on Google's list numbers. It does not say Sume lists these three Gemini models at these prices; the catalog line is the authority.

For an exact number, multiply the catalog cost_usd by the count you plan to submit. Failed or cancelled image generations are not billed, and a client that disconnects early is treated as a failed generation, so the arithmetic above is the worst case for a batch that completes.

What does Sume not do here?

Sume does not expose Google's token rates or let you price an image in tokens. You get one USD figure per image and the model list shows what each entry accepts. If a Gemini tier is not in the catalog for your workspace, it is not available through Sume, whatever Google's page says.

To compare tiers, pull the model list, sort by cost_usd, and test the two cheapest candidates on your own prompt. Resolution and the number of reference images move the price more than the logo on the tier.

How do I keep an image batch inside a budget?

Fix the size first, then the tier. Decide the largest dimension you ship and generate at that size or slightly above. Then test the two cheapest tiers on your own prompt and keep the cheaper one if the difference does not show on your layout.

Submit with idempotency keys so a retry does not create a second charge, and read the usage ledger after a trial batch to confirm the real per-image amount. The ledger is the check on every number in this post, including the arithmetic table.

If you need the exact Sume rate for a particular model, the catalog line is the answer. The Google figures here are for orientation only.