Vercel's pricing page counts usage in two dozen separate meters

The workload

Vercel publishes two self-serve tiers and a negotiated one. The pricing page lists roughly two dozen separately billed units – edge requests, data transfer, function invocations, Fluid Active CPU, image transformations and more – each with its own allotment and overage rate. The limits page adds a second, unpriced layer: hard ceilings on projects, deployments per day and function duration that apply regardless of price. Neither document computes a bill for a specific app; that arithmetic is left to the account holder.

What the documents show

Verified: as retrieved 16 September 2026, Hobby costs $0 per month and Pro costs $20 per month per seat, per the pricing page. Hobby includes 1 million edge requests and 100 GB of Fast Data Transfer monthly; Pro includes 10 million edge requests and 1 TB, with overage from $2 per additional million requests and $0.15 per additional GB. Verified: the limits page, last updated 3 September 2026, adds ceilings the pricing page omits. Hobby functions default to a 10-second execution window with a 60-second maximum for projects not using Fluid compute; Pro's default is 15 seconds with a 5-minute maximum. Hobby is also capped at 200 projects and 100 deployments per day, with no overage price for either figure. The two documents do not contradict each other; one names dollars, the other names walls money cannot move.

The operating cost

Verified: entry cost is $0 per month on Hobby, bounded by the ceilings above. Pro costs $20 per month per seat before usage overage, per the pricing page, which states starting rates of $0.60 per million function invocations and $0.128 per hour of Fluid Active CPU. Estimated: neither document can produce a dollar total for a specific workload, since that needs the workload's own request volume and compute time; any total is the account holder's own monthly calculation. Named add-ons include $20 per month for an additional developer seat.

The stop condition

Neither document names a spending ceiling, only technical ones. This is an editorial stop condition, not a vendor-stated one: a reasonable point to reconsider architecture is when overage recurs for three consecutive months and approaches the size of the base subscription, since the marginal unit price is then doing more work than the plan itself. A vendor-stated stop point is structural: Hobby's 100-deployments-per-day and 200-project ceilings have no overage price at all, so a workload reaching them must move to Pro or stop.

• Which of the pricing page's two dozen billed units does this product actually consume in volume, and which sit unused?
• Does the workload's peak function duration fit inside the plan's default timeout, or does it depend on a higher maximum that must be configured?
• Over three months, does overage cost more than upgrading a seat or changing the architecture would?

A pricing page and a limits page from the same vendor describe two different constraints: one priced, one absolute. Both are dated to the day they were retrieved because both can change without notice.