Statsig

Statsig is a unified product-development platform that bundles feature flags ("feature gates"), A/B testing/experimentation, product analytics, and session replay into one system on a single event stream. Founded by ex-Facebook engineers who built Facebook's internal experimentation tooling, it positions itself as a more analytics-native, statistically rigorous alternative to pure flag tools like LaunchDarkly and Flagsmith. The defining commercial decision is that feature-flag and dynamic-config checks are unlimited and free on every tier, you are billed instead on metered "events" (analytics/experiment exposures, session replays). In May 2025 OpenAI acquired Statsig, with the founder becoming OpenAI's CTO of Applications, which has raised both confidence (deep resourcing) and some uncertainty among buyers about long-term independence.

For feature flags specifically, Statsig is strong: 30+ open-source SDKs across server, client, and edge; local/in-process evaluation so gate checks add effectively no network latency; edge deployment via Vercel Edge Config (cited ~15ms P99), Cloudflare Workers, Fastly Compute and Akamai; sophisticated targeting rules, environments, progressive/scheduled rollouts, and automatic rollback on negative metric impact. Because every flag can have metrics attached, teams get "measured rollouts", lightweight A/B tests on any release, without standing up separate tooling. The platform claims trillions of daily flag checks across billions of users, and its public status page shows 99.96–100% 90-day uptime across config-delivery and regional endpoints.

The trade-offs are real. The pricing model rewards flag-only usage but can produce large, surprising bills once you lean into analytics, experimentation, or session replay at high traffic (Reddit anecdotes of ~$40k/yr enterprise quotes circulate). There is no self-hosted/on-prem option (warehouse-native deployment is the closest, Enterprise-only), which rules it out for strict data-sovereignty needs, and the statistical engine is somewhat of a black box for data-science teams that want to validate p-value math. Reviewers consistently praise ease of setup and the consolidated UI but flag engineering-skewed documentation and a learning curve for non-engineers and for getting full value out of experimentation. Net: an excellent choice for product/eng teams that want flags plus experimentation plus analytics in one place, less ideal for teams needing self-hosting, transparent stats internals, or strict per-event cost predictability at scale.