After generate_image on sume/auto, why can't the agent name the model?

The agent cannot name the model because Sume does not say. For sume/auto, Sume selects the family for you and never discloses which one ran. The job's model field stays sume/auto, and GET /v1/images/models does not list it. If a report must name the model, pin a catalog id instead of leaving payload.model blank on generate_image.

What the docs say, row by row

Sume's Image generation page has a Sume specifics table. The behavior that matters for an agent report is below.

When blank is right, and when to pin

The MCP docs say to omit payload.model and route to sume/auto unless the user named a family. That is the right default for a quick draft where any good result will do. Pin a model when you need one of these:

• A repeatable look across a series: Auto may choose differently next time.
• A named model in a client report or a style guide.
• A feature only some models have, such as many reference images; for example, gpt-image-2.5 takes up to 16 references.
• A price you can calculate in advance, from the catalog entry.

How to pin from an agent

Ask the agent to call image-models_list and read the capabilities of the entries, then image-models_get for the one it plans to use. The ids come from the live catalog, not from memory, because the catalog changes. Then it submits generate_image with that id in payload.model.

A good instruction reads: list the image models, choose one that supports references, show me its id and estimated price with dry_run=true, and wait for my yes before submitting. With Write granted, the paid call also needs an idempotency_key.

What the agent can say safely

After an Auto run, the safe statement is: generated with Sume Auto, job id and request id, billed amount. It should not guess the family from the look of the result. For the cost, quote the cost or the debited amount from usage_get, not the held estimate. If the user later wants the same style, the instruction is to rerun with a pinned model and compare.

Same rule for video

The same behavior exists for video. The Video generation page says sume/auto is an addition that only Sume has, and that the poll response reports sume/auto without disclosing which family ran. An agent that generates both images and clips on Auto should therefore label its outputs with the job id and the billed cost, which are always known, and not with a model name.

This is also why a style guide that says "made with model X" cannot rely on Auto. If your brand needs the claim, pin the id in the instruction and have the agent copy it into the report from the catalog entry, not from the picture.