Wan 3.0 has no fine-tuning or batch mode; what Sume requests allow

On Sume you cannot batch, fine-tune or pass provider options to Wan 3.0, and Alibaba's own documentation says the model does not support batch inference, fine-tuning or function calling either (read 2026-10-04). A request that tries to tune the provider is refused with a 400.

What you can do is submit many independent jobs, with one idempotency key per job.

What does Alibaba say Wan 3.0 does not support?

Alibaba's Model Studio page for Wan 3.0, updated September 28, 2026, lists three unsupported features: batch inference, fine-tuning and function calling. It also gives a rate limit of 300 requests per minute across regions and a 30-second maximum clip.

What happens if you send provider options?

Sume's video router runs one backend per model in v1, so there is nothing to forward to. A non-empty provider.options returns 400 unsupported_parameter, and allowed_passthrough_parameters is an empty list for every model. An empty options object is accepted.

The same strictness applies to seed and size. No v1 model accepts them, and the request is rejected instead of ignoring the field.

How do you run a lot of Wan jobs without batch?

Submit jobs one by one, each with its own Idempotency-Key, and poll them. A replay of the same key returns the original job, so a network retry does not create a second charge.

Because Alibaba documents a limit of 300 requests per minute, keep any bulk submit loop well below that if you also call Alibaba directly. For Sume requests, read the response status and back off on any 429.

• One request per clip; no list endpoint to submit many at once.
• Stable key per clip, such as a SKU id plus a version number.
• Poll polling_url; read usage.cost once the job is completed.

What does a safe bulk loop look like?

A small loop with one key per item.

import os, requests

H = {"Authorization": f"Bearer {os.environ['SUME_API_KEY']}"}
items = {"sku-101": "A ceramic mug rotating on a white table",
         "sku-102": "A leather wallet opening on a desk"}
for sku, prompt in items.items():
    r = requests.post(
        "https://api.sume.com/v1/videos",
        headers={**H, "Idempotency-Key": f"{sku}-v1"},
        json={"model": "wan-3.0", "prompt": prompt, "duration": 5,
              "resolution": "480p"},
    )
    print(sku, r.status_code, r.json().get("id"))