Softwr

Testing · head to head

GrowthBook vs Split

GrowthBook logo

GrowthBook

Testing

Open source feature flags and A/B testing that run experiment analysis inside your own data warehouse

From
Free
Rated
-
Full reviewWebsite
Split logo

Split

Testing

Feature flags tied to a metrics pipeline that attributes production impact to each release, now Harness FME

From
On request
Rated
-
Full reviewWebsite

The short version

  • Only GrowthBook has a free tier, so it costs nothing to try first.
  • Each has a real cost: GrowthBook warehouse-native analysis is only an advantage if you have a warehouse with clean event data in it; without one, GrowthBook has nothing to compute against and you are back to buying an event-collecting platform.; Split split is now a module of the Harness platform and split.io redirects to harness.io, so buying it means entering a Harness commercial relationship rather than a standalone flag vendor contract.
  • They diverge on capability: GrowthBook covers Feature flags, Split covers Feature flags with targeting.
  • Prices and features above were last checked on 31 August 2026.

Where they differ

Only the attributes on which GrowthBook and Split actually diverge.

Attributes where GrowthBook and Split differ
AttributeGrowthBookSplit
Starting priceFreeOn request
Pricing modelPer user per monthquote
Free tierYesNo
PlatformsWeb, Self-hosted, DockerWeb, API, iOS, Android, Java, Node.js, Python, Go, .NET

Identical on both: user rating (Not yet rated), category (Testing).

What each one covers

Drawn from each product's published feature list. An absence here means we hold no record of it - not that the product lacks it.

Only in GrowthBook

  • Feature flags
  • Warehouse-native analysis
  • Reusable metric definitions
  • Bayesian and frequentist engines
  • CUPED variance reduction
  • Visual editor
  • Self-hosting

Only in Split

  • Feature flags with targeting
  • Metrics impact engine
  • Statistical significance testing
  • Event ingestion
  • SDKs and streaming updates
  • Audit and approvals
  • Harness platform integration

Both cover

  • Guardrail metrics

What people use each for

The jobs each tool is most often brought in to do.

GrowthBook

  • An engineering team that wants feature flags without a per-monthly-active-user bill that grows with product successnot Split
  • A data team that needs experiment results to reconcile exactly with the metrics reported in its BI toolnot Split
  • A company with data residency or privacy constraints that cannot send user-level event data to a third-party vendornot Split
  • Running long-tail experiments where sequential testing and CUPED shorten the time to a decision on low-traffic surfacesnot Split

Split

  • A product organisation that wants every release measured against conversion and latency by default, not only deliberate experimentsnot GrowthBook
  • Running A/B tests with real statistical significance rather than comparing two dashboard linesnot GrowthBook
  • Catching a regression at five per cent rollout because a guardrail metric fires before the change reaches everyonenot GrowthBook
  • A team already committed to Harness for CI and CD that wants flags in the same platform and contractnot GrowthBook

Where each one falls short

Documented limitations, not opinions. Every one is a constraint you would hit in normal use.

GrowthBook

  • Warehouse-native analysis is only an advantage if you have a warehouse with clean event data in it; without one, GrowthBook has nothing to compute against and you are back to buying an event-collecting platform.
  • Metrics are SQL, so a marketing or product team without data engineering support cannot define or fix them, and a badly written metric silently produces a confident wrong answer.
  • Cloud Pro bills per seat and then adds CDN request and bandwidth overages, so a high-traffic site can find the variable component exceeds the seat cost it budgeted for.
  • Self-hosting means you own the Mongo database, the SDK endpoint availability and the upgrade path, and a flag service that is down is a production incident, not an analytics inconvenience.
  • The statistics engine exposes real choices, Bayesian versus frequentist, sequential testing, variance reduction, and the tool will not stop a team that configures them wrongly, so it rewards statistical literacy and punishes its absence.

Split

  • Split is now a module of the Harness platform and split.io redirects to harness.io, so buying it means entering a Harness commercial relationship rather than a standalone flag vendor contract.
  • Harness no longer publishes per-module prices, so an organisation that wants only feature management cannot get a list price and must negotiate a module out of a platform bundle.
  • The measurement engine needs traffic volume to reach significance, so a product with modest usage gets flags with an experimentation layer it will rarely be able to conclude anything from.
  • Metric attribution depends on correctly instrumented events flowing into Split, which is an integration project in its own right; without it you have paid for an experimentation platform and are using it as a flag switch.
  • As part of a larger DevOps suite the roadmap now competes with CI, CD, security and cloud cost modules for attention, so feature management is no longer the company’s single focus the way it was as an independent product.

Pricing, plan by plan

GrowthBook

Free
  • Open source (self-hosted)Free
    • Unlimited users, feature flags and experiments
    • One project
    • MIT licensed, you run the infrastructure
  • Starter (cloud)Free
    • Up to 3 users
    • 1 project
    • Unlimited flags, experiments and traffic
  • Pro (cloud)$40/month
    • Per seat, up to 50 users
    • 3 projects
    • CDN overage at $10 per million requests beyond the included allowance
  • Enterprise$undefined/year
    • Unlimited users and projects
    • SSO, audit logs and advanced permissions
    • Available on cloud or self-hosted

Split

On request
  • Harness FreeFree
    • Harness Open Source and community support
    • Intended for individuals and small teams
    • Feature management capability limited at this tier
  • Harness Essentials$undefined/year
    • Bundled DevOps modules including CI, CD, IaC and security testing
    • Standard support
    • Pricing requires contacting sales
  • Harness Enterprise$undefined/year
    • Mix and match from 15 or more modules including Feature Management and Experimentation
    • Premier support with a dedicated account manager
    • Higher pipeline concurrency and organisation limits

Which should you pick?

Choose GrowthBook if

  • You need feature flags.
  • You want to start without paying.
  • You work on Web, Self-hosted, Docker.
  • You also want warehouse-native analysis.
GrowthBook review

Choose Split if

  • You need feature flags with targeting.
  • You work on Web, API, iOS, Android, Java, Node.js, Python, Go, .NET.
  • You also want metrics impact engine.
Split review

Questions people ask

Is GrowthBook or Split better?
Neither clearly leads. GrowthBook starts at Free and Split at On request, and user ratings are close enough to be indistinguishable. Choose on capability and platform support.
Which is cheaper, GrowthBook or Split?
GrowthBook has a free tier; the other does not. Paid plans start at Free for GrowthBook and On request for Split.
Does GrowthBook or Split run on more platforms?
GrowthBook runs on Web, Self-hosted, Docker. Split runs on Web, API, iOS, Android, Java, Node.js, Python, Go, .NET.
Can I use GrowthBook for free?
Yes. GrowthBook has a free tier, so you can try it without paying. Split starts at On request.
What is GrowthBook best used for?
GrowthBook is most often used for an engineering team that wants feature flags without a per-monthly-active-user bill that grows with product success, a data team that needs experiment results to reconcile exactly with the metrics reported in its bi tool, a company with data residency or privacy constraints that cannot send user-level event data to a third-party vendor, running long-tail experiments where sequential testing and cuped shorten the time to a decision on low-traffic surfaces. Of those, an engineering team that wants feature flags without a per-monthly-active-user bill that grows with product success and a data team that needs experiment results to reconcile exactly with the metrics reported in its bi tool are not what Split is typically brought in for.
What can GrowthBook do that Split cannot?
GrowthBook covers Feature flags, Warehouse-native analysis, Reusable metric definitions, Bayesian and frequentist engines. Split covers Feature flags with targeting, Metrics impact engine, Statistical significance testing, Event ingestion. Both handle Guardrail metrics.

Answered from the vendors’ own pages

GrowthBook: Is the self-hosted version actually usable, or crippled?

It is genuinely usable: unlimited users, flags and experiments on one project under an MIT licence. The paid tiers add multiple projects, SSO, audit logs and support.

Split: Is Split still called Split?

It is now sold as Harness Feature Management and Experimentation after the Harness acquisition, and split.io redirects to harness.io.

GrowthBook: Do I need a data warehouse?

For experiment analysis, effectively yes. Feature flagging works without one, but the reason to choose GrowthBook over a flag-only tool is the warehouse-native statistics.

Split: Can I buy just feature flags?

Not from a price list. Harness publishes bundle tiers and directs single-module buyers to sales.

GrowthBook: How does the cost compare with LaunchDarkly or Optimizely?

GrowthBook charges per seat rather than per monthly active user or per tracked event, so cost stops scaling with traffic. At high volume that difference is large.

Split: What does it do that a plain flag service does not?

It attributes changes in your product metrics to specific flags with statistical significance testing and alerts on guardrail metrics.

GrowthBook: Does it support mobile and server-side flags?

Yes, there are SDKs across JavaScript, React, Python, Go, Ruby, PHP, Java, Kotlin, Swift and more, covering client, server and mobile evaluation.

Split: Do we need a lot of traffic?

Yes. Statistical tests need volume to conclude, so low-traffic products get limited value from the measurement layer.

Share

Related pages

More on GrowthBook

More on Split

Keep looking

Other head to heads