One provider behind Hy Image 3.5 Preview vs Sume's one image endpoint

The short answer

Both services run Hy-style image calls through a single route, but for different reasons. OpenRouter's listing for Hy Image 3.5 Preview says the model is hosted by one provider and that OpenRouter forwards every request to it directly (read 2026-10-09). Sume's Image API publishes one endpoint, sume, for each catalog model. Sume does not list Hy Image 3.5 Preview, so there is no Hy route to compare on Sume.

The two routing statements

OpenRouter names the provider, Tencent Cloud, and shows its own latency and availability numbers. Sume says it does not disclose the upstream provider identity, and that the multi-provider routing fields exist in the schema but mostly have no effect in v1.

What this means for fallbacks

With one provider behind a model, a provider outage is a model outage. The OpenRouter page shows 91.57% three-day availability for Hy Image 3.5 Preview, against 99.92% uptime. On Sume you cannot add a second provider for the same model, because the routing fields have no effect. The fallback you can build is a second model: if the first one fails, send the same request to another Sume image row.

Check that the second model accepts every parameter you send. A parameter that a model does not list fails with 400 unsupported_parameter, so a fallback needs a request body made for it, not a copy of the first one.

What to take from it

A single provider means that when that provider has an outage, nothing else on that listing can serve the request. The availability figure of 91.57% comes from that arrangement.

A Sume call goes to Sume's endpoint, and the docs describe job envelopes and retries, so your own code deals with one interface, but the model behind it still has its own limits.

A short checklist

Read the live list with GET /v1/images/models and store the model ids you use. For each, read the endpoints route for supported_parameters and pricing. Keep a primary and a backup model per job type, and log which one served each image. Watch the Sume docs and catalog for a Hy row if you want that model through one API.

For a team that wants two providers behind one name, the choice is to have the failover in your own code or to use a gateway that does it. Sume's image docs describe the job envelope and the model list, and your client code can treat each model id as a separate row with its own price.

Whatever you pick, log the model id, the status and the cost for every call. When a listing shows low availability, that log is the evidence you will need to decide whether to switch.