Grep for nano-banana-2 in your code: those calls bill as 2.1

If your code still sends google/nano-banana-2 or nano-banana-2 to Sume, it keeps working: the Image API docs say those ids run as Nano Banana 2.1 (google/nano-banana-2.1), and the job stores the 2.1 id. The catalog no longer lists the retired id. At the 1K default Nano Banana 2.1 is $0.08 list, which is 10 cents billed.

What the docs say

The retirement note is in the Sume specifics table of the Image API docs: Nano Banana 2 is retired; google/nano-banana-2 and nano-banana-2 still work and run as Nano Banana 2.1, and the job stores the 2.1 id. The catalog does not list the retired id.

That means a model name in your logs may not match what ran. If you log job.model from the job record, you will see the 2.1 id, not the one you sent.

What to change in your code

Update the id to google/nano-banana-2.1 so your logs, dashboards and cost reports name the model that actually ran. The old id is a compatibility path, not a listed catalog row.

If you have tests that assert on the model string in the response, check them: model echoes the id you requested on the synchronous response, while the job record stores 2.1.

• Search your repository for nano-banana-2 without a .1.
• Update cost tables that priced the old model.
• Add a test that reads the job record and checks the stored id.
• Keep a note of the date you changed it.

Price effect

Because the request runs as 2.1, you pay the 2.1 rate. At 1K that is 10 cents; the catalog scales admission for 0.5K, 2K and 4K, so check usage.cost if you ask for 4K. The Pro row is a different price: 19 cents at 1K. Do not confuse Pro with the retired alias.

A grep to run today

Search your repository, your prompt library and any no-code automations for the strings nano-banana-2 and google/nano-banana-2. Anything that matches without .1 is still riding the compatibility path. Replace it, redeploy, and confirm in one job record that the stored model is the 2.1 id. Do the same in saved Formats or workflow definitions if you keep model ids there, since those copies do not update themselves.

Why a silent alias is still worth fixing

An alias hides change. If Sume later changes how the legacy id is handled, code that names the retired model is the code most likely to surprise you. A current id in the request also makes bills, dashboards and support questions line up with what actually ran.

The same advice applies to the bare Image Router ids such as gpt-image-2 and nano-banana-2.1: the docs say Sume accepts them as aliases for their org/slug equivalents. Prefer the full form, for example google/nano-banana-2.1 and openai/gpt-image-2, in new code.