Decision supported

feature flag pricing comparison

Which feature flag platform is cheapest at your specific architecture, once each vendor's published price is converted into the unit you will actually consume.

01

Published prices, read on the same day

Every figure in the table below was read from the vendor's own pricing page on 8 September 2026. Nothing here is a rumoured price, an analyst estimate, or a number carried over from an older comparison. Where a vendor does not publish a figure, the table says so rather than guessing.

One vendor is worth calling out before the table. Split, now sold as Harness Feature Management and Experimentation, no longer publishes a price for the module. It appears on the Harness pricing page as a module within the Enterprise plan, with no tier, no rate, and no free allowance shown. If you are comparing against an older article that quotes a Split price, that price is no longer published.

Published pricing read from each vendor's own pricing page on 8 September 2026. Confirm before purchase; packaging changes frequently.
VendorFree tierEntry paid priceSelf-hosting
LaunchDarklyDeveloper: $0, unlimited seats, unlimited flags, 5 service connections, 1K client-side MAU/moFoundation: pay as you go after a 14 day trial. $10 per service connection/mo, and $8.33 per 1K client-side MAU/mo billed yearlyNot published
StatsigDeveloper: $0, 2M events/mo, unlimited flag and config checks, 50K session replays/moPro: $150/mo, 5M events/mo, then $0.05 per 1K eventsWarehouse native deployment on Enterprise
FlagsmithFree: $0, 1 team member, 50K API requests/mo, 1 project, unlimited environmentsStart-Up: $40/mo billed yearly or $45/mo monthly. 3 members, up to 1M API requests/moYes. Open source, plus self-hosted or private cloud on Enterprise
UnleashOpen source: free, no seat limit, 1 project, 2 environments, 5,000 flags per instancePay-as-you-go: $75/seat/mo, unlimited projects and environments, 53M API requests/mo then $5 per millionYes. Open source, plus self-hosted or hybrid on Enterprise with a 5 seat minimum
ConfigCatForever Free: $0, unlimited seats, 5M config JSON downloads/mo, 10 flags, 2 environments, 2 productsPro: $110/mo, 25M config JSON downloads/mo, 100 flags, 3 environments, 3 productsDedicated tier published at $4,500/mo
GrowthBookStarter: free cloud for up to 3 users, unlimited flags and experiments. Open source self-hosted is free with unlimited seatsPro: $40/seat/mo, up to 50 users, plus $10 per additional million CDN requests and $1 per additional GBYes. Open source, plus a self-hosted Enterprise tier
PostHog1M feature flag requests/mo includedPay as you go above the free allowance. The per-request rate is not published on the feature flags pricing pageNot covered on the feature flags pricing page
Harness FME (formerly Split)Not publishedNot published, contact salesEnterprise, contact sales
02

The comparison problem: eight vendors, seven different meters

A price comparison is only meaningful when the things being compared are measured the same way. In this category they are not. Eight vendors bill on at least seven distinct units, and two of those units are properties of your deployment topology rather than of your usage.

This is why cross-vendor tables in listicles are close to useless for forecasting. A table that puts $75 per seat next to $150 per month next to pay-as-you-go is comparing a headcount number, a subscription, and a formula. The cheapest vendor for a small team serving millions of anonymous users is not the cheapest vendor for a large team serving a few hundred enterprise tenants, and no single ranking can be correct for both.

The unit each vendor actually bills on, as published on 8 September 2026.
VendorPrimary billing unitSeats billed?
LaunchDarklyServer-side SDK instances multiplied by environments, plus unique client-side context keysNo
StatsigAnalytics events. Flag and config checks are explicitly not billedNo
FlagsmithTeam members, plus monthly flag-evaluation API requestsYes
UnleashSeats, with an included API request ceiling above which requests are meteredYes
ConfigCatConfig JSON downloads, plus flag, environment, and product countsNo. Seats are unlimited on every tier
GrowthBookSeats, with CDN request and bandwidth overageYes
PostHogFeature flag requestsNo
Harness FMENot publishedNot published
03

How to make the numbers comparable

Measure your own five numbers first, then run each vendor's published formula against them. This takes an afternoon and it is the only way to get a defensible answer.

Do the arithmetic for two scenarios: today, and your own twelve-month forecast. Several of these meters grow against changes that have nothing to do with the flag product. A move to autoscaling multiplies LaunchDarkly service connections. A mobile launch multiplies any client-side or request meter. Hiring multiplies the seat-priced vendors and leaves the others untouched.

  • Peak concurrent server-side SDK instances, multiplied by the environments each connects to.
  • Unique end users evaluated client-side per month, split into signed-in, anonymous, and synthetic.
  • Total flag evaluation requests per month, which is the meter for Flagsmith, Unleash, and PostHog.
  • Analytics or experiment events per month, which is the meter for Statsig.
  • Number of people who need write access, which is the meter for Unleash, GrowthBook, and Flagsmith.
04

Where an open source option genuinely changes the answer

Three of these products have a real open source edition rather than a marketing gesture. Unleash publishes an open source instance with no seat limit, capped at 1 project, 2 environments, and 5,000 flags. Flagsmith publishes an open source edition and offers self-hosted or private cloud on Enterprise. GrowthBook publishes an open source self-hosted edition with unlimited seats.

The saving is real, and so is the cost you take on. A self-hosted flag service becomes a tier-zero dependency: if it is down or stale, your application either fails open into an unintended state or fails closed into an outage. Budget for the on-call rota, the upgrade path, and the availability target before treating the licence saving as a saving.

The seat-priced vendors are where self-hosting pays back fastest, because the licence cost scales with a number you cannot easily control. Where the meter is requests or events, self-hosting moves the cost into your own infrastructure rather than removing it.

  • Unleash open source: no seat limit, 1 project, 2 environments, 5,000 flags per instance.
  • Flagsmith: open source edition, with self-hosted and private cloud on the Enterprise tier.
  • GrowthBook: open source self-hosted edition with unlimited seats.
  • ConfigCat: seats are unlimited on every hosted tier, which removes the usual self-host motive.
  • Statsig: warehouse native deployment on Enterprise, which keeps event data in your own warehouse.
05

The cost that none of these tables contains

Every figure above prices the delivery of a flag. None of them prices the flag that nobody can remove. A flag left live because no one can prove which customers depend on it costs engineering time on every release, blocks packaging changes, and quietly becomes a contractual commitment nobody signed.

That cost does not appear on any vendor's meter, and switching vendors does not reduce it. It is a record-keeping problem: which accounts evaluate this flag, how much recurring revenue sits behind them, who approved keeping it, and when that decision is next reviewed.

Before optimising a subscription, count the flags in your account that have outlived their release. If that number is large, the subscription is not the expensive part.

Product status and boundary

Varistra is in development, and this page is useful without it

Varistra is not generally available. There is no signup, no trial, and no published price, so nothing on this page depends on buying anything. What Varistra is being built to do is inventory feature flags and other customer-specific product behavior and connect them to account, revenue, usage, support, maintenance, and decision evidence. It will not evaluate runtime flags, ship an SDK, enforce infrastructure state, calculate a complete customer P&L, or make roadmap and commercial decisions.

Evidence base

Sources and further reading

Buyer questions

What teams ask before they choose a control layer

Which feature flag tool is cheapest?

There is no single answer, because the vendors bill on different units. A small team serving millions of anonymous users is often cheapest on a seat-priced vendor such as Unleash or GrowthBook. A large team serving a small number of tenants is often cheapest on a request or event meter such as ConfigCat or Statsig. Run your own numbers through each published formula.

How much does Split.io cost?

As of 8 September 2026, Harness does not publish a price for Feature Management and Experimentation, the product formerly sold as Split.io. It appears as a module within the Enterprise plan with no tier, rate, or free allowance shown. Older articles quoting a Split price are quoting a price that is no longer published.

Are there genuinely free feature flag tools?

Yes. Unleash, Flagsmith, and GrowthBook publish open source editions you can self-host. ConfigCat, Statsig, LaunchDarkly, and PostHog all publish free hosted tiers with usage caps. The binding constraint differs by vendor: seats for Flagsmith, flag count for ConfigCat, service connections and client-side MAU for LaunchDarkly.

Why do feature flag vendors bill on such different units?

Because they started from different products. Statsig prices analytics events because it began as an experimentation platform. LaunchDarkly prices SDK connections and client-side contexts because it began as delivery infrastructure. ConfigCat prices config downloads because it is a configuration delivery service. The meter reflects the vendor's original centre of gravity, not a shared industry standard.

Does switching vendors reduce feature flag technical debt?

Not on its own. Lifecycle tooling helps engineering find flags that look ready for removal, but a flag stays live when the blocker is commercial rather than technical: nobody can prove which customers depend on it or what revenue is attached. That evidence sits outside the delivery platform.