GrowthBook

GrowthBook is an open-source, warehouse-native platform that combines feature flags, A/B testing and product analytics in a single tool that you can either self-host for free or run on GrowthBook Cloud. We rate it 84/100 — if you already trust your data warehouse and you want a serious experimentation engine without LaunchDarkly-tier pricing, GrowthBook is one of the best deals in the category.

What is GrowthBook?

GrowthBook was founded in 2020 by Jeremy Dorn (CTO) and Graham McNicoll (CEO, previously CTO at Education.com), went through Y Combinator's Winter 2022 batch, and raised a Series A on June 17, 2025, bringing total funding to roughly $23.1M from investors including Khosla Ventures and Y Combinator. As of May 7, 2026, the growthbook/growthbook repository has crossed 7,700 GitHub stars, and the company says GrowthBook is now used by 3,000+ companies.

The thesis is simple: most teams already have a clean source of truth for their metrics — their data warehouse — so an experimentation tool should query that warehouse directly instead of forcing teams to re-emit events into a vendor's pipeline. GrowthBook plugs into BigQuery, Snowflake, Databricks, Redshift, ClickHouse, Postgres and seven more sources, runs SQL-defined metrics on your own data, and never copies that data out. The latest stable release, v4.3.0, shipped on February 4, 2026.

Key Features of GrowthBook

• Warehouse-native experimentation: 11 supported data sources including BigQuery, Snowflake, Databricks, Redshift, ClickHouse and Postgres — experiment data never leaves your warehouse.
• Serious stats engine: CUPED variance reduction, sequential testing, Bayesian and frequentist analyses, sample-ratio-mismatch (SRM) checks, post-stratification and contextual bandits all in the box.
• 24 SDKs: Official client libraries for React, Vue, Next.js, Node, Python, Go, Java, PHP, Ruby, Kotlin/Android, Swift/iOS, Flutter, Edge SDKs and more — flag evaluation runs locally for sub-millisecond decisions.
• SQL-backed metric definitions: Conversion rates, ratios, quantiles and revenue metrics are defined as plain SQL templates that anyone on the data team can audit.
• Built-in product analytics: Dashboards, funnels and ad-hoc charts share the same metric definitions used by experiments — one source of truth, not two.
• MCP server and full REST API: The 2026 releases shipped an MCP server that lets AI assistants create flags, start experiments and clean up stale rollouts directly, in addition to the existing webhook + REST surface.

What Users Say About GrowthBook

On G2 and Product Hunt, the recurring praise is around three things: the open-source license (teams can fork or self-host with zero per-seat tax), the warehouse-native architecture (no need to ship sensitive product events to a third party), and unusually fast hands-on support from the founding team. The recurring complaints are equally consistent: documentation is dense and gets lost in cross-references, the UI can feel intimidating to non-technical PMs, and SDK setup — especially around streaming vs. fetch-and-cache configurations — takes more reading than most teams expect on day one. Hacker News threads tend to compare GrowthBook favourably against LaunchDarkly on cost (multiple commenters report 5x savings after migrating) but note that LaunchDarkly's enterprise compliance and observability integrations remain ahead.

GrowthBook Pricing

GrowthBook offers a generous free tier on cloud and unlimited free self-hosting, with paid tiers gated mostly on stats features and team-management controls.

Plan	Price	Key Limits
Self-Hosted Open Source	$0	Unlimited users, unlimited flags and experiments, MIT-licensed core
Cloud Starter	$0	Up to 3 users, unlimited flags and experiments on GrowthBook Cloud
Cloud Pro	$40/user/month	Up to 50 users; unlocks CUPED, sequential testing, visual editor, advanced stats
Enterprise	Contact sales	Guardrail metrics, holdouts, custom roles, SSO, audit logs, dedicated support

Who Should Use GrowthBook?

Best for: Engineering and product teams that already use a modern data warehouse, want their experimentation platform to read from it directly, and either need to self-host for compliance reasons or are tired of paying $75+/seat to LaunchDarkly. It is also a strong choice for solo developers and small startups, because the Cloud Starter and self-hosted tiers are genuinely free and feature-complete.

Not ideal for: Non-technical product teams that want a one-click visual editor and zero SQL, very large enterprises that need deep ServiceNow / Terraform / Datadog automation, or teams without a data warehouse at all.

Pros and Cons

Pros:

• Free, unlimited self-hosting under MIT — the Open Core boundary is unusually clearly documented.
• Warehouse-native: your event data never has to be re-piped into a vendor.
• Stats engine is best-in-class for the price — CUPED, Bayesian, sequential and SRM checks are not paywalled at enterprise tier.
• Founders are unusually responsive on Slack, GitHub issues and Product Hunt according to multiple G2 reviews.
• MCP server (2026) makes flag and experiment management possible directly from AI coding assistants.

Cons:

• Documentation is comprehensive but hard to navigate; multiple G2 reviewers cite it as the steepest part of the learning curve.
• SDK bundles can be larger than competitors — reported as a real issue in low-bandwidth or edge contexts.
• UI is engineer-first; PMs and marketers may need a guided tour.
• Enterprise compliance integrations (Terraform, ServiceNow, broad SOC2/ISO frameworks) are still less mature than LaunchDarkly's.

Alternatives to GrowthBook

The closest commercial competitor is LaunchDarkly — richer enterprise plumbing and streaming-first SDKs, but several times the seat cost and no full self-host option. Statsig sits in the same warehouse-native experimentation lane and has aggressive free pricing for startups, while PostHog bundles feature flags and experiments inside a broader product analytics suite that may double as your replacement for Mixpanel. Flagsmith and Unleash are honest open-source feature-flag-only alternatives if you don't need experimentation at all.

Verdict: Is GrowthBook Worth It?

GrowthBook is the right pick when you already invest in a modern data stack and your bottleneck is an honest experimentation engine that doesn't charge per active user. Solo developers and bootstrapped startups can live on the free tier indefinitely. Mid-sized teams will find the $40/user Pro tier dramatically cheaper than LaunchDarkly while keeping a more credible stats engine. Where GrowthBook still loses ground is in enterprise compliance plumbing and in handholding for non-technical users — if either of those is your top concern, you'll want to evaluate LaunchDarkly or Statsig in parallel. We rate it 84/100 for being the best price-to-power ratio in the warehouse-native experimentation category in 2026.

Frequently Asked Questions

Is GrowthBook free?

Yes. Self-hosting GrowthBook is free and unlimited under the MIT license, and GrowthBook Cloud has a free Starter plan for up to 3 users with unlimited feature flags and experiments. Paid Pro starts at $40/user/month and unlocks advanced stats and the visual editor.

Is GrowthBook open source?

GrowthBook is Open Core. The bulk of the codebase is MIT licensed and lives at github.com/growthbook/growthbook; a small set of enterprise directories are governed by a separate commercial license, and that boundary is documented in the repo's LICENSE file.

What data warehouses does GrowthBook support?

11 sources as of v4.3.0, including BigQuery, Snowflake, Databricks, Redshift, ClickHouse, Postgres, MySQL, Athena, MS SQL Server, Presto/Trino and Mixpanel. Metrics are defined as SQL templates that run inside your warehouse.

How does GrowthBook compare to LaunchDarkly?

GrowthBook is materially cheaper (Pro at $40/user vs LaunchDarkly's enterprise pricing), open-source, fully self-hostable, and includes a real experimentation stats engine in the box. LaunchDarkly leads on streaming-first SDK behaviour, deeper enterprise compliance integrations and observability automation. Teams optimising for cost and data sovereignty pick GrowthBook; teams optimising for enterprise procurement pick LaunchDarkly.

Does GrowthBook have an MCP server?

Yes. The 2026 releases ship an MCP server that lets AI coding assistants create feature flags, kick off experiments and retire stale flags directly via the Model Context Protocol.