All posts
November 20, 2026·9 min read

Semrush API pricing: what a million units really cost

Gaurav Singh
Gaurav Singh
Founder, ad-vertly
Semrush API pricing: what a million units really cost

Semrush API pricing has two meters. The plan that unlocks the API is Advanced, and since September 2026 it's quote only. The units are published: $100 for one million a month, falling to $65 per million on the twenty million package. One million units buys 100,000 lines of live organic keyword data.

The Semrush API bill has two meters

One meter is the subscription that grants access at all. The other is the unit pool your calls draw from. Semrush sells them separately, and the second one is where the bill actually moves.

The developer documentation is blunt about the sequencing. Upgrade to the plan that permits API access, and the number of units will still be zero. Access and consumption are two purchases. You buy the right to make calls, then you buy the calls.

That split matters more when an agent holds the key. A person runs a report and reads it. An agent loops, retries on a timeout, and pulls a full keyword set to answer a question a person would have settled with a filter. The seat count stays flat while the balance drains.

Semrush's own knowledge base, checked on 20 November 2026, publishes a package table: 1M units at $100 a month, 5M at $450, 10M at $750, and 20M at $1,300. Units renew on your billing date. Unused units expire at the start of the next cycle, so a quiet month isn't a credit.

There's one free allowance in there as well. Operational calls such as creating or deleting an API key consume no units at all. Your spend is data, not administration.

What one million API units costs in 2026

Divide price by units and the ladder stops looking flat:

  • 1M units at $100 a month works out to $100.00 per million.
  • 5M units at $450 works out to $90.00 per million.
  • 10M units at $750 works out to $75.00 per million.
  • 20M units at $1,300 works out to $65.00 per million.

That's a 35% spread between the smallest and largest package for an identical unit. Most pages on this query quote $100 per million and stop, which is the entry rate and the worst one on the list.

The trap in the other direction is buying the big package to chase the cheaper rate. At 20M units you pay $1,300 whether you consume the balance or not, and the balance does not roll forward. A 20M package you half use is $1,300 for 10M units, which is $130 per million. Worse than the 1M tier.

Size the package to your heaviest realistic month, and treat the rate discount as a bonus for volume you already have.

How much does one report line cost?

Units aren't charged per call. They are charged per line of data returned, and the rate varies by report. The one Semrush documents with a worked example is the Domain Organic Search Keywords report: 10 API units per line for live results, 50 per line for historical data.

So a unit package converts into a line count:

  • At 10 units a line, the 1M package buys 100,000 lines and the 20M package buys 2,000,000.
  • At 50 units a line, the same two packages buy 20,000 lines and 400,000.

In dollars, live keyword data runs $0.00100 per line at the entry package and $0.00065 at the top one. Historical keyword data runs $0.00500 and $0.00325. The five times multiplier on history does more damage to a budget than the package choice ever will.

Rates differ across the rest of the API, and Semrush prints each report's cost next to the report itself in the documentation. Read the rate for the endpoints you actually plan to call. A line count from one report does not transfer to another, and the spread between the cheapest and the most expensive report is wide enough to change which package you need.

What a real keyword-tracking job burns in units

Abstract rates get clearer with one realistic workload. Say you track 500 keywords across 20 competitor domains, weekly, pulling live organic positions.

That's 10,000 lines per pull, so 100,000 units. Four and a bit pulls a month comes to roughly 434,800 units, which fits the 1M package at $100 a month with room to spare. Two pulls a week still fits.

Now add twelve months of history to each pull so you can show movement. Historical data costs five times the live rate, so the same 10,000 lines become 500,000 units instead of 100,000. Each pull is now 600,000 units, and a month at four pulls is 2,400,000 units. That forces the 5M package at $450.

One extra column turned a $100 month into a $450 month. It's a 4.5x rise for a comparison field, and nobody budgets for it because the line count in the request never changed.

The practical rule is three steps: units per pull is lines times the report rate, multiply by pulls per month, then round up to the next package. Do that before you write the integration, not after the first invoice.

Which plan actually unlocks the API?

This is where Semrush's own pages disagree, and it's worth arriving with both versions in hand.

The developer documentation for API access says Standard API access is available to users of the SEO Toolkit who hold a Business subscription, and the knowledge base repeats that framing. The pricing page, read on 20 November 2026, lists no plan called Business at all. The tiers are SEO at $139 a month, Starter at $199, Pro+ at $299, and Advanced, which is the one naming API access and is priced Custom behind a Talk to sales button.

So the honest reading is that the top tier carries the API, the tier got renamed, and the price left the public page. Confirm the tier name and the number in writing on your order form. A support article naming a plan that no longer exists isn't a quote.

There's a second surprise on that pricing page. MCP access now appears on Starter at $199, which is the cheaper way to let an assistant query Semrush data in conversation rather than pulling rows into your own store. If that covers the job, you can skip the unit meter entirely, and for most agent setups it's the better first purchase.

The plan side of that decision, including the annual discount and the per-seat charges, is in the semrush pricing breakdown and the annual vs monthly comparison.

Why your first month usually lands on the wrong package

Three mechanics push a new integration onto the wrong tier, and none of them show up during a trial.

The first is the zero balance after upgrade. You have access, the key generates, and every call fails until a package is bought. Teams that test with a single request and then ship a backfill discover the real burn rate in production.

The second is expiry. Because unused units vanish at renewal, an oversized package is a recurring loss rather than a one-off. You pay for the ceiling, not the usage.

The third is that a backfill isn't a steady month. Loading twelve months of history across 20 domains is a one-time job with a unit cost that looks nothing like your maintenance load. Price it on its own and buy units for it on its own, or it will eat the package you sized for ongoing work.

The cost drivers behind the floor price are worth reading alongside this, especially the crawl economics in why Semrush costs what it does.

How do you keep an agent under its unit ceiling?

An agent spends units faster than a person doing the same job, because it has no instinct for when a report is already big enough. Four controls do most of the work.

  • Cap the response size. Reports that support the display_limit parameter return only the top N lines, and you pay per line, so the parameter is a spending control as much as a formatting one.
  • Cache what you already pulled. A keyword set fetched this morning does not need fetching again this afternoon, and a local store costs nothing per read.
  • Kill retry storms. An agent that retries a timed-out request five times pays five times. Bound the retries and let it fail loudly.
  • Check the balance before a big job. Remaining units sit in Subscription info, and a run that starts near zero fails halfway through and leaves partial data behind.

One quiet saver sits on top of those. Asking for 5,000 lines on a domain holding 40 keywords still sizes the request for 5,000. Set limits from what you expect, not from the maximum the endpoint allows.

When the API is the wrong call

The API is priced for volume, and its floor is a quote-only enterprise tier. If your real need is a handful of competitor keyword checks a week, you're buying a metered data pipe to do a lookup.

Run the arithmetic first. If your monthly pull lands under a few hundred thousand units you're in the $100 band, and that's a fair price for automated access. If the need is conversational, MCP on the Starter tier gets an assistant talking to Semrush data without a unit package at all.

Two companion pieces are worth the detour. If you're wiring any of these into an agent, the Search Console API row limit explains why response caps break integrations, and Mailchimp API keys covers the credential handling side. For a rival pricing model to sanity-check per-line rates against, Serpstat pricing runs a different structure entirely.

The short version: the plan is a floor you can't see, the units are a meter you can read, and the meter is the only part worth optimising.