Claude Message Batches queue limits by tier vs Sume bulk runs

A Claude batch queue holds 200,000 requests on the Start tier, 300,000 on Build and 500,000 on Scale, while one Sume bulk run takes 1 to 100 items with 1 to 16 running at once. They solve different jobs. Claude's batches are a discounted queue for text requests. A Sume bulk run is a window of paid media renders you watch.

What are the Claude numbers?

The batch processing page and the rate limits page give the figures.

What else does the batch page promise?

The page lists a 50% cost discount, a 256 MB size cap, expiry after 24 hours, and a note that most batches finish within an hour. Results come back per request as succeeded, errored, canceled or expired, and only succeeded requests are billed.

How does a Sume bulk run compare?

A bulk run creates up to 100 Format runs from one POST. Each item has the same body as a single run, and concurrency from 1 to 16 sets how many run at once. A malformed item returns 400 invalid_request with details.index before any queue exists, so you fix it before anything is billed.

Sume does not offer a batch discount that I could find in the docs, and the queue has no webhook, no list endpoint and no cancel-queue endpoint. You cancel a child run with POST /v1/format-runs/{run_id}/cancel.

• Creating a queue spends the write budget: Free 120, Pro 300, Startup 600, Scale 1200 per minute.
• Queue status is queued, running or completed. Completed means every item is terminal, not that every item succeeded.
• Branch on counts.failed and counts.canceled.

Which should I use?

For tens of thousands of text requests where a day's wait is fine, the Claude queue is built for it. For a hundred renders where each one has a spend cap and a receipt, use a bulk run, and split larger jobs into several queues with distinct idempotency keys. See the bulk runs reference.