Replicate cost estimate before running vs Sume dry_run preview

Replicate points you to the model's own page for a cost estimate. Sume gives you a preview in the request path: call a paid MCP tool with dry_run=true and it returns an admission and cost preview without submitting the job. Add max_spend_usd to cap a real submit.

Sources: Replicate's pricing page and Sume's MCP tools and gates, read 2026-10-01.

Where does Replicate show an estimate?

The pricing page says you will find estimates for how much any model will cost on the model's page. It also says most models are billed by the time they take to run on the hardware in use, and some by input and output. So the estimate lives per model, outside the call you make.

What does dry_run do on Sume?

The gates table defines dry_run=true as "Admission/cost preview only; do not submit the job." The docs also say to prefer generation_admission_preview and/or dry_run before expensive bursts, and that max_spend_usd is enforced only when provided.

What happens when I do submit?

Generation admission says paid generation uses public Sume USD estimates: the estimate is reserved when the request is accepted, and successful completion captures the reserved usage. For worked examples see estimate AI video cost before you run.

Which one should I use before a burst?

For a single create, just submit; the docs say ordinary single creates do not need admission theater. Before dozens of paid calls, run one dry_run=true per distinct request shape and set max_spend_usd on the real calls.