Metabase Pricing: Metabase Cost for Starter, Pro and Enterprise, With the Per-User Math

Metabase pricing, as published on metabase.com on 10 September 2026: the open source edition is free to self-host, Starter is $100 a month including 5 users and then $6 per extra user, Pro is $575 a month including 10 users and then $12 per extra user, and Enterprise starts at $20,000 a year. Paying yearly takes 10 percent off every line, so the plans become $1,080 and $6,210 a year, and each extra user $65 or $130 a year.

That is the whole rate card, and it is simpler than most BI pricing. The surprises are underneath it: who counts as a user, how the AI and transform meters bill, and the headcount where the Pro plan costs as much as Enterprise. Every figure here comes from Metabase's pricing page or its billing documentation, read the same day, or is plain arithmetic on those figures.

Metabase pricing plans at a glance

Both paid plans carry a 14-day free trial. Dashboards, questions and data sources are not metered: Metabase answers its own FAQ on this plainly, saying it "costs the same no matter what you make, or how much you explore." The bill moves with people, and with three optional usage meters covered further down.

How much does Metabase cost per user?

Because each plan bundles a block of users, the price per person falls as the team grows. On yearly billing:

Starter drops below $8 a person past 25 users. Pro only gets under $20 at 50. The gap between them is almost entirely the $475 difference in the monthly base, so the question for a small team is simple: do you need row-level permissions, SSO or customer-facing embedding this year? If not, Starter is the plan, and the jump to Pro is roughly $5,800 a year for 25 people.

Where Pro stops making sense: the 117-user line

Pro on yearly billing is $6,210 plus $130 for every user past ten. That reaches $19,990 at 116 users and $20,120 at 117, which is past the published $20,000 Enterprise floor. Metabase is unusually candid about what the extra money buys. Asked the difference between Pro and Enterprise, its pricing page answers "Feature-wise, zero" and adds that "it's the same software". Enterprise adds people and paperwork: a dedicated success engineer, a 1-day support SLA, help with procurement, and air-gapped deployment for an extra fee.

So a team heading past about 115 users on Pro should ask for an Enterprise quote before renewal. At that size the list price is the same, and Enterprise pricing is described as "custom pricing aligned with your business model", which is the one line on the page that invites negotiation.

Five things on the Metabase pricing page that most guides get wrong

1. The Enterprise floor is $20,000 a year

The pricing page reads "Price starts at $20,000 / year". At least one guide on page one of Google still puts the Enterprise entry point at about $15,000 and lists Pro at $500 a month plus $10 a user, both older figures. Budget from the $15,000 number and you start the sales conversation a quarter short.

2. Storage starts at $2 per million rows

Storage is optional and only matters if you upload spreadsheets or CSVs rather than connecting a database. The page says pricing "starts at $2 for 1M rows of storage" and that "1M stored rows are always free". Figures of $40 per 500,000 rows still circulate on page one; that is not what metabase.com says today.

3. The AI meter is described two ways

Metabase's AI features (Metabot, AI-assisted SQL and the MCP server) are included in every plan, open source too. Most teams bring their own model provider's API key, in which case, per the billing docs, "you pay your provider directly for token usage, and Metabase adds nothing on top." If you use Metabase's managed AI service instead, the pricing page quotes $3.75 per million tokens with the first million included, while the billing documentation describes the same service as "a flat 20% markup on the underlying token cost, with your first 1M tokens each month free." Those are two different formulas for one meter. If AI usage is a real budget line for you, get the rate confirmed in writing.

4. The page lists two prices for advanced transforms

Transforms include 1,000 runs on Starter and Pro, then $0.01 a run. For advanced transforms, the same section of the page says both "$0.02 per run" and "$0.01 per run". We counted every dollar amount in the page markup to check: there are exactly fifteen distinct figures ($0.01, $0.02, $2, $3.75, $6, $12, $65, $90, $100, $130, $517.50, $575, $1,080, $6,210 and $20,000), and both transform prices are among them. Self-hosting sidesteps the question, because transforms run on your own infrastructure and are unlimited.

5. An unused account is a paid account

This is the one that costs real money. The billing documentation says that "if an account exists, but has not been deactivated, that account will count toward your bill, even if no one signs in and uses that account." Deactivated accounts are charged by the day up to deactivation. On annual plans the upfront payment and true-ups are non-refundable, and "if your user account number dips below the amount you've already paid for, you won't be refunded." Put deactivation in your offboarding checklist, next to revoking the laptop.

Does Metabase charge for viewers?

Yes. There are no viewer-only seats: "Everyone who signs into your Metabase (logging in directly or via SSO) is a user for purposes of billing." That extends to embedding. Full app embedding and modular embedding require viewers to sign in, so they count as users. Guest embedding does not require a sign-in and does not add users. If you plan to put dashboards in front of customers, which embedding mode you use changes the bill more than any plan choice.

Is Metabase free?

The open source edition is free as software, and for a technical team it is a genuinely capable tool. What it costs is operations. Someone provisions the server, runs the upgrades, backs up the application database and keeps it patched. Somebody also has to notice when it falls over at 7am before the Monday metrics review, which is what uptime monitoring for the Metabase URL is for. Those costs are real but they are yours to size. If nobody on the team wants to own a server, Starter at $1,080 a year is cheaper than an afternoon a month of an engineer's time.

Metabase vs Tableau on price

Take 25 people: 3 who build, 6 who edit and 16 who only read. Tableau Standard, at the role prices Tableau publishes ($75 Creator, $42 Explorer, $15 Viewer a month, billed annually), costs $717 a month or $8,604 a year. Metabase Starter costs $2,380 a year for the same 25 people and Pro costs $8,160. Tableau is the deeper visualization tool, but on price alone Metabase Starter is roughly a quarter of it. Our Tableau pricing breakdown has the full role math, and Metabase vs Tableau compares the two on more than price.

When a question tool is the cheaper answer

Metabase is a BI tool. You get a visual query builder, dashboards, permissions and embedding, and the value comes from someone building those dashboards and keeping them current. Before pricing it, check whether that is actually the problem you have. Often the real problem is narrower: people keep asking the one engineer who knows SQL for numbers that no existing dashboard shows.

That second problem is what we built Agentsql for. It connects read-only to Postgres, MySQL, Snowflake or BigQuery, takes a question in plain English, writes and runs the SQL, and returns a chart and a one-line answer with the query shown. There are no dashboards to build first. Starter is $39 a month billed yearly for one database connection, and Team is $99 for five connections and up to 10 seats, with no cap on questions. For ten people that is $1,188 a year against $1,405 on Metabase Starter, so the prices are close. The deciding question is whether anyone will build and maintain dashboards. If yes, Metabase is the better buy. If the questions change every week, a question tool fits better. The Metabase alternative page lays out that trade honestly, and if your data sits in Postgres, the Postgres query generator page shows exactly what a generated query has to get right there.