Hy Image 3.5 lists 13 ratios; GPT Image 2.5 on Sume lacks six

Tencent's Hy Image 3.5 Preview lists 13 aspect ratios on OpenRouter, and six of them are not in the GPT Image 2.5 row on Sume: 1:2, 1:4, 2:1, 2:3, 3:2 and 4:1. Sume does not list Hy Image, so if you need those ratios on Sume, use a row that lists them. Nano Banana 2.1 lists 1:4 and 4:1; Nano Banana Pro lists 3:2 and 2:3; Qwen Image and FLUX.2 pro list 1:2 and 2:1.

The two lists

Where each missing ratio lives on Sume

These come from the Sume catalog rows, not from Tencent. Each model accepts only the values its descriptors list, so read supported_parameters before pinning one.

Why the count matters less than your target

A ratio list is a menu, not a quality score. Thirteen ratios help when you produce many placements from one prompt: a 4:5 feed post, a 9:16 story, a 4:1 header, a 2:3 poster. If you only publish at 1:1, 16:9 and 9:16, the GPT Image 2.5 row already covers you and the extra ratios change nothing.

The practical question is which of your placements fall outside the list. Write them down, check each against the model you want, and split the job: one model for the ratios it lists, a second for the rest. Because sume/auto hides the family it chose, pin an id whenever the ratio matters.

A short checklist before you pin a ratio

Five checks keep a ratio mismatch from reaching a paid call.

• Open the models endpoint and read aspect_ratio.values for the exact id you will call.
• If the ratio is missing, decide whether a crop from a listed ratio is acceptable for the placement.
• For edits, send aspect_ratio: "auto" to match the reference; omitting the field is not the same as auto.
• Pin the model id when the ratio matters, because sume/auto does not disclose which family ran.
• Record the ratio in your job metadata so a later audit can see what you asked for; Sume stores it on the job and does not send it to the provider.

What to do with a banner or tall strip

For a 4:1 banner there is one Sume row, Nano Banana 2.1. For a 1:2 tall strip you have several. If your target is 3:2 photography and you prefer GPT Image 2.5's text handling, render at 4:3 or 5:4 and crop; the Image API cannot return a ratio the row does not list.

A request with a ratio the row does not list is rejected, so a small pre-check against the models endpoint saves a failed call. Failed generations are not billed.

Source of the lists

The Hy Image ratio list is the one shown on OpenRouter's model page on 2026-10-08, and it is the list for that provider's endpoint; Tencent Cloud's own API may differ. The Sume lists are the catalog descriptors for each row on the same date. Compare like with like only after you have read the endpoint you intend to call.