A feature flag starts life as an if statement and a config value. Then a team has forty of them, three environments, a product manager who wants to turn something on for ten percent of users on Tuesday, and an engineer who left last spring without documenting which flags are safe to delete. That is the point at which people go looking for feature flag software: a control panel that decides who sees what, writes down who changed it, and keeps old flags from rotting in the codebase. This ranking is written for engineering teams of roughly five to a few hundred developers. We compared how long a first flag takes to reach production, what the bill looks like when traffic grows, whether flag definitions can leave with you, and how much of the stack you could still run yourself if the vendor changed course. The order is the verdict.
Visibility in this ranking can be paid for. Payment moves a vendor's position within the
shortlist; it never adds a vendor, and it never changes a word of the review. The largest vendors in feature flag software
cannot hold places 1 to 3. How it works: placement disclosure ·
editorial process.
How we ranked these
Setup: how fast the first flag reaches production
The first flag is easy in every product here; the second week is where they differ. We looked at how many SDKs cover your languages, whether server-side evaluation happens locally or by a network call on each check, and what happens when the flag service is unreachable. A client-side SDK that blocks page render on a remote call is a different risk from a server SDK that keeps a cached ruleset in memory. We also counted the steps to a working local environment for self-hosted options, since one Docker command and a Postgres dependency is a very different afternoon from a Kubernetes chart. Support for the OpenFeature standard counted in a product's favor because it keeps your application code from being rewritten around a single vendor's client.
The real price: seats, requests and what gets metered
Vendors in this market meter four different things: seats, API requests, monthly active users and tracked events. A free plan that is generous on one axis can be tight on another. ConfigCat, for example, lets the whole company in but caps a free account at ten flags; Flagsmith allows unlimited flags but one team member. We priced a hypothetical team of eight developers with a mobile app, a web app and roughly ten million flag evaluations a month, and noted which meter would grow fastest. Per-seat plans punish teams that want product managers and support staff to flip flags themselves. Usage plans punish chatty client SDKs. Where a price is quote-only, we say so rather than guess at a figure.
Getting your data out: flag definitions and evaluation history
What you accumulate in a flag platform is a set of targeting rules, segments, environment overrides and an audit trail of who changed what. Moving that to another product is the real exit test, and the answer depends on whether a documented API or export covers all of it, not only the flag names. We checked for REST or GraphQL management APIs, bulk export, Terraform providers and Git-based storage. Products that keep flags as files in a repository (or can mirror to one) score well here, because the history is yours and diffable. Experimentation data adds a second question: whether exposure and result data sit in your own warehouse or only in the vendor's analytics, since a lost experiment history cannot be reconstructed from flag configs.
Independence from the vendor: licenses, hosting and ownership
A feature flag service sits in the request path of your product, so a pricing change or an acquisition lands in production. We read each license and each recent ownership change as part of the score. Open-source cores that can be run on your own servers (BSD, MIT or AGPL, each with different obligations) give you an exit that does not depend on the vendor's goodwill. Hosted-only products depend entirely on the vendor's uptime and roadmap. Ownership matters too: DevCycle is now a Dynatrace subsidiary and Statsig sits inside Amplitude, which may suit some buyers and unsettle others. Data residency counted as well, with EU region options and private cloud deployments weighed for European teams.
Who it's for: engineering teams that ship weekly or faster
The audience here is a product engineering team that releases often enough that coupling deploys to launches hurts: a startup moving to trunk-based development, a mid-sized SaaS company that wants a kill switch before every big launch, a platform team serving a dozen squads. It also covers teams that want to add A/B testing on top of flags without buying a separate experimentation suite. It is not written for a company that deploys quarterly and needs a plain config file, nor for marketers who want to test headlines on a landing page; that job belongs to website personalization tools. And a single developer with one environment can get a long way with an environment variable and a restart.
The 11 tools, reviewed
#1 Flagsmith
Open-source feature flags and remote config, hosted or self-hosted · United Kingdom · flagsmith.com
Open source Remote config Private cloud
Flagsmith earns the top place because it removes most of the usual trade-offs. The code is BSD-3-Clause, so a team can run it in its own cluster and keep the data. The hosted plans publish prices, with a free tier that allows unlimited flags, environments and segments. Hosting is where it stands out: shared SaaS across eight regions, fully managed private cloud in a chosen region, or self-hosted. It blends flags with remote config values, which suits teams that would otherwise keep two systems. A London-registered company also answers the data-location question for many European buyers.
Where it falls short
The free plan allows one team member and 50,000 requests a month, so any real team is on a paid plan immediately. SAML single sign-on and governance features start at the Scale-Up tier at $250 a month. A/B testing arrives as an add-on rather than the core of the product, and experiment analysis is shallower than what GrowthBook or Statsig offer.
Wrong for
A data science-led team whose main need is rigorous experiment statistics should look at GrowthBook or Statsig, where flags are a means to measurement rather than the other way round.
Pricing: Free plan with 50,000 requests a month and one team member; Start-Up from $40 a month billed yearly for 1,000,000 requests; Scale-Up from $250 a month billed yearly; Enterprise quoted. (free plan)
Visit Flagsmith →
#2 Unleash
Open-source feature management built around activation strategies · Norway · getunleash.io
AGPL license Self-hosted Nordic vendor
Unleash sits second because its open-source edition is a serious product, not a teaser: gradual rollouts, kill switches, stale flag tracking and a long list of official SDKs, free to self-host under AGPL-3.0. The paid cloud plan is easy to reason about, one price per seat with a large request allowance, so a chatty client app does not produce a surprise bill. It is a Norwegian company with an Oslo address, which helps European procurement. Unleash Edge exists to scale evaluation for front-end clients without hammering the main server, and that matters once traffic grows.
Where it falls short
The free open-source instance is limited to one project, two environments and 5,000 flags, and it omits single sign-on, SCIM and granular role-based access control. The cloud plan costs $75 per seat, which hurts when product and support staff need accounts. Self-hosted use requires at least five seats on the paid plan, and the AGPL license will worry some legal teams.
Wrong for
A company that wants a large team, including non-engineers, to manage flags on the cheap should compare ConfigCat's unlimited-seat plans or PostHog's seatless usage pricing.
Pricing: Pay-as-you-go at $75 per seat a month with a 14-day trial, including 53 million API requests a month and $5 per million overage on cloud; enterprise by annual contract. (open source)
Visit Unleash →
#3 GrowthBook
Open-source flags and warehouse-native experimentation · United States · growthbook.io
Experimentation Warehouse-native Open core
GrowthBook is the choice when a flag exists to answer a question. Experiment results are computed from your own warehouse (Snowflake, BigQuery, Redshift and others), so the history stays in systems you control, and the Pro plan adds sequential testing, CUPED variance reduction and sticky bucketing. The repository is open core with an MIT-licensed base, and the cloud Starter plan has unlimited flags and experiments. At $40 per seat on Pro it is cheaper per head than several rivals. It earns third place, not higher, because it is strongest for teams that already have a warehouse and an analytics habit.
Where it falls short
Value depends on having a data warehouse with clean event data; without one, the analysis side has little to work on, although a managed warehouse option exists with overage at $30 per million events. Approval workflows, audit log export and SSO sit in the Enterprise tier. Starter support is community-only, and the free plan stops at three users.
Wrong for
A small engineering team that just wants kill switches and percentage rollouts, with no experiments planned, will be paying for analysis it does not use; Flagsmith or ConfigCat fits that job better.
Pricing: Starter is free for up to 3 users and 1 project; Pro is $40 per seat a month for up to 30 users and 3 projects; Enterprise is quoted; the open-source self-hosted edition is free. (free plan)
Visit GrowthBook →
#4 LaunchDarkly
Enterprise feature management with flags, experiments and AI configs · United States · launchdarkly.com
Incumbent Usage-based Enterprise governance
LaunchDarkly is the reference product for the category, and an honest ranking has to say so. The free Developer plan has unlimited seats and unlimited flags, the SDK coverage is wide at 30, and the product has grown into session replay, error monitoring and AI configuration management around the flag core. Large organizations with many squads, compliance review and a need for approval flows tend to land here. It sits fourth because its two meters, service connections and client-side monthly active users, grow with success, and because a smaller team gets more per dollar from the products above.
Where it falls short
Costs scale with two usage meters, and a mobile app with a large user base makes the client-side MAU line the biggest part of the bill. The free plan keeps data for only 14 days. Hosting is a service you depend on, with no self-hosted edition mentioned on its pricing page, and the breadth of modules can feel heavy for a team that wants only flags.
Wrong for
A small team, or one that must run everything inside its own infrastructure, should start with Flagsmith or Unleash, which publish simpler prices and offer self-hosting.
Pricing: Free Developer plan with unlimited seats; Foundation is pay-as-you-go at $10 per service connection a month and $8.33 per 1,000 client-side monthly active users billed yearly; Enterprise is quoted. (free plan)
Visit LaunchDarkly →
#5 ConfigCat
Flat-priced feature flags with unlimited seats · Hungary · configcat.com
Unlimited seats EU vendor Flat pricing
ConfigCat is the easiest of these to budget. Plans are flat monthly fees, and every plan includes unlimited seats, service connections, monthly active users and API access, so the bill does not move when a product manager joins or traffic doubles. The company is Hungarian, states GDPR compliance, and lists ISO 27001:2022 certification across all plans, which short-circuits a security questionnaire. The paid tiers are inexpensive relative to usage-metered rivals. Its position reflects scope: it handles flags and targeting well, but it does not try to be an experimentation platform.
Where it falls short
The free plan allows only 10 flags, 2 environments and 7 days of audit log, so even a small product outgrows it quickly. Paid tiers are separated by limits on flags, environments and audit retention rather than by workflow features. Experimentation and statistics are not the focus, and the on-premise option is described only briefly on the pricing page.
Wrong for
A team that wants flags tied tightly to experiment results and metric analysis should look at GrowthBook or Statsig instead.
Pricing: Forever Free at $0 with 10 flags; Pro $110, Smart $325, Enterprise $900 and Dedicated $4,500 a month, with annual billing saving up to 10 percent; unlimited seats and MAUs on every plan. (free plan)
Visit ConfigCat →
#6 PostHog
Feature flags inside an all-in-one product analytics suite · United States · posthog.com
Bundled analytics Seatless Usage-based
PostHog wins a place for teams that want flags next to the analytics that show what the flag did. Pricing is simple to model: the first million flag requests each month are free, then a declining per-request rate, with no charge per seat, so everyone in the company can have an account. Flags and experiments share that free allowance. Because the flag, the funnel and the replay live in one product, engineers spend less time wiring event data to a separate experiment tool. It lands sixth because flags are one product among many there, and the specialist tools offer deeper release workflows.
Where it falls short
Flag management is one module in a large suite, so release-specific governance such as approvals and scheduled changes is less central than in dedicated tools. Request-based billing means a polling client SDK can run up charges without anyone noticing. Buying flags usually means adopting the analytics side too, which is a larger commitment than some teams want.
Wrong for
A platform team that wants a standalone, self-contained flag service with deep governance across dozens of squads should look at LaunchDarkly or Harness instead.
Pricing: 1,000,000 flag requests free every month, then $0.000100 per request falling to $0.000010 at high volume; no per-seat charges. (free plan)
Visit PostHog →
#7 Statsig
Flags, experiments and analytics on one event stream · United States · statsig.com
Experimentation Unlimited checks Part of Amplitude
Statsig takes the unusual step of not charging for flag checks at all; it meters events, so a team can evaluate flags freely and pay only for logged exposures and metrics. The free Developer tier includes 2 million events and 50,000 session replays a month, and Pro starts at $150. The platform combines flags, experimentation, analytics and replay, and offers warehouse-native deployment for Enterprise. It is placed seventh since the company now operates under Amplitude's name on its terms, which changes the long-term picture, and because a flags-only buyer pays for a suite.
Where it falls short
Event-based metering is harder to predict than a seat count, because exposures, custom events and ingested metrics all count. Pro allows only one project per subscription, and multiple projects need Enterprise. The product is hosted, with no self-hosted flag service described. Its recent move into Amplitude leaves roadmap and pricing direction less certain than with an independent vendor.
Wrong for
A buyer who requires an independent vendor, an open-source core, or a strictly self-hosted flag service should look at Unleash or Flagsmith.
Pricing: Developer plan free with 2 million events a month and unlimited flag and config checks; Pro $150 a month with 5 million events then $0.05 per 1,000 extra events; Enterprise quoted. (free plan)
Visit Statsig →
#8 Optimizely
Feature management inside an enterprise experimentation suite · United States · optimizely.com
Quote only Experimentation Enterprise
Optimizely brings a mature experimentation business to flags: targeted delivery by user, region or attribute, progressive rollouts from 1 percent upward, instant kill switches and flag lifecycle governance. The company says its 20-plus SDKs evaluate with sub-millisecond local decisions across backend, mobile and edge, and the SDKs are published as open source on GitHub. It suits an enterprise that already runs Optimizely for web experimentation or content and wants one vendor. It sits eighth because a small team cannot see a price, start free or self-host, so evaluation takes a sales conversation.
Where it falls short
No public pricing or free tier on the product page we read, so budgeting means a sales call. The page gives no hosting detail. The product is positioned for enterprise experimentation programs, and teams that only need dependable flags get a heavier commercial process than the self-serve alternatives require.
Wrong for
A startup or mid-sized team that wants to sign up, ship a flag and see a price the same afternoon should look at Flagsmith, ConfigCat or PostHog.
Pricing: No prices or plans shown on the product page; access is through sales, a demo request or a get-started form. (none)
Visit Optimizely →
#9 Harness Feature Management & Experimentation
Flags and experiments built into a software delivery platform · United States · harness.io
CI/CD integrated Modular Quote only
Harness offers feature flags as part of a delivery platform, which makes sense when releases already run through its pipelines: rollouts, release monitoring that detects impact, and rollback live next to the deployment tooling. The product page lists percentage-based targeting, A/B testing with warehouse-native experimentation, dynamic configs, RBAC and audit trails, with SDKs for JavaScript, Node.js, Python, Java, Go, .NET, Ruby, PHP and Flutter. A free tier exists for individuals and small teams. It sits ninth since pricing beyond the free tier requires a conversation, and the value is highest for existing Harness customers.
Where it falls short
Module-level prices for flags are not published, so comparing cost with competitors is hard. The page does not describe hosting options or self-hosting. Buying into the module means buying into a platform vendor, and teams that use another CI/CD system lose most of the integration advantage.
Wrong for
A team on a different delivery toolchain that wants a lightweight, transparent price should consider ConfigCat or Flagsmith rather than a platform module.
Pricing: A free tier for individual developers and small teams; Essentials and Enterprise plans are quoted through sales, and Feature Management & Experimentation appears as a module in the modular Enterprise plan. (free plan)
Visit Harness Feature Management & Experimentation →
#10 DevCycle
OpenFeature-native flag management on an edge network · United States · devcycle.com
OpenFeature-native Edge network Dynatrace
DevCycle calls itself the first OpenFeature-native platform, and its design shows it: 15-plus SDKs, an edge network with automatic fallbacks, and a free plan with unlimited seats and flags. Pricing counts client-side MAUs, config requests and tracked events rather than people, so a whole company can log in. It also lets agents manage flags through an MCP integration. It lands tenth because DevCycle is now part of Dynatrace following an asset acquisition, effective January 13, 2026, and independent buyers should expect product direction to follow Dynatrace's observability priorities.
Where it falls short
The free plan covers only 1,000 client-side MAUs, and audit logging plus roles and permissions begin at the $500-a-month Business plan. Four usage meters (MAUs, cloud config requests, server config requests and events) are harder to forecast than a flat fee. The ownership transition is recent, and the legal entity named on its terms is in flux.
Wrong for
A buyer who wants a stable, independent vendor with a self-hostable option should pick Flagsmith or Unleash, and a very small mobile app on a tight budget will find ConfigCat cheaper.
Pricing: Free plan with 1,000 client-side MAUs; Business $500 a month billed annually ($625 monthly) for 100,000 MAUs; Enterprise quoted; no seat charges. (free plan)
Visit DevCycle →
#11 FeatBit
Open-source feature flags with a free self-hosted core · United States · featbit.co
Open source Self-hosted OpenFeature
FeatBit is the budget route for teams that want to own the whole stack. The Foundation tier is described as full-featured open source with unlimited seats, users, projects and service connections, and deployment targets include Kubernetes, AWS, Azure and Docker Compose. Paid support starts at $3,999 a year, a fraction of what usage-metered cloud products cost at scale. It supports OpenFeature, offers a CLI and an MCP server for AI coding agents, and comes with open-source server and client SDKs. It ranks last because it is newer and smaller than the rest.
Where it falls short
The vendor is much smaller than its rivals, so expect a thinner ecosystem of integrations, less public track record and fewer third-party guides. Advanced role-based access control costs extra even on the Enterprise tier. Support responses are listed at 12 business hours on Enterprise, and the privacy page gives no public street address for the company.
Wrong for
A buyer who needs a large vendor with an established customer base and global support coverage should choose LaunchDarkly or Flagsmith instead.
Pricing: Foundation tier is free forever for self-hosting; Enterprise is $3,999 a year, with advanced RBAC at $600 a year extra; Enterprise Premium is quoted; unlimited seats and projects. (open source)
Visit FeatBit →
What the data says about this market
ICT service exports grew from 9.1 percent of world service exports in 2013 to 14.48 percent in 2023 in the World Bank series (indicator BX.GSR.CCIS.ZS), and the European Union went from 10.58 percent to 17.55 percent. Software that ships continuously, rather than once a quarter, is part of that shift, and feature flags are one of the practices that make continuous shipping tolerable: code can reach production dark and be switched on later by someone other than the person who wrote it. The market has moved accordingly, from homegrown config tables to dedicated services with audit trails and access controls.
Of the 11 vendors ranked here, eight are headquartered in the United States, and the other three are in the United Kingdom, Norway and Hungary. Four publish a self-hostable edition that costs nothing to run: Flagsmith, Unleash, GrowthBook and FeatBit. Eight offer a free tier or free plan in their hosted product (Flagsmith, GrowthBook, ConfigCat, LaunchDarkly, PostHog, Statsig, DevCycle and Harness), while Optimizely published no price or plan on the product page we read. Only two of the eleven, Unleash and GrowthBook, state a flat per-seat price for their paid cloud plan; the rest meter requests, users, events or flag connections.
The pattern to watch is absorption. Experimentation vendors have folded flags into broader platforms, analytics tools have added flags as a feature, and two products in this list changed hands recently: DevCycle moved to Dynatrace in January 2026, and Statsig now carries Amplitude's name on its legal terms. Meanwhile the OpenFeature specification gives buyers a way to keep application code vendor-neutral. A flag service you can swap with a configuration change is a better bet than one whose client library is woven through every service, whichever company owns it next year.
The 11 ranked vendors, counted
- Headquarters by region: North America 8, Europe 3
- By country: United States 8, Hungary 1, Norway 1, United Kingdom 1
- Pricing model: Usage-based 4, Open source + paid tiers 2, Per seat / month 2, Quote only 2, Flat monthly 1
- Free option: Free plan 8, Open source 2, None 1
Counted from the 11 vendors on this page. More in our market data.
For the wider market behind feature flag software, read our report The Global Shift to ICT Services,
or browse all industry reports.
Questions and answers
What is the best feature flag software in 2026?
Flagsmith takes first place in our ranking because it combines a permissively licensed open-source core, a hosted plan with a published price from $40 a month on annual billing, and hosting choices that include eight cloud regions, private cloud and self-hosting. Unleash is the better pick for teams that want a Nordic-built, self-hosted system with a flat per-seat price, and GrowthBook suits teams whose flags exist mainly to run experiments against their own data warehouse. LaunchDarkly has the broadest feature set but ranks lower because of its size and metered pricing.
What is the difference between a feature flag and a configuration setting?
A configuration setting is read at startup and applies to everyone; changing it usually means a restart or redeploy. A feature flag is evaluated per request or per user at runtime, using targeting rules, so two users can get different answers at the same moment. Flags also tend to be temporary: a release flag should be deleted once the rollout is complete, while a config value may live for years. Some products, such as Flagsmith, deliberately blend the two by offering remote config values alongside flags.
Should we build feature flags ourselves instead of buying software?
A homegrown table of flags is fine for a handful of on/off switches and one team. It stops being fine when you need percentage rollouts that stay sticky per user, audit history, role-based permissions, scheduled changes and SDKs in more than one language. A self-hosted open-source product such as Unleash or Flagsmith is usually a better middle path than writing your own, because you get those features while keeping the data and the runtime inside your infrastructure. Building from scratch rarely saves money once maintenance is counted.
Do feature flags slow down our application?
Not noticeably, if evaluation is local. Server-side SDKs generally download the ruleset and evaluate it in memory, so a flag check is a function call. Trouble comes from client-side SDKs that wait on a network response before rendering, or from services that call a remote API on every check. Ask each vendor whether the SDK caches rules, what the default value is when the service is unreachable, and whether an edge or relay proxy is available. DevCycle describes an edge architecture for this reason, and Unleash offers an Edge component for front-end clients.
Is a free plan enough to run feature flags in production?
For a small product, often yes, but read the meter. Flagsmith's free plan covers 50,000 requests a month and one team member. PostHog gives 1,000,000 flag requests free each month. Statsig's free tier counts 2 million events monthly with unlimited flag checks, and LaunchDarkly's free Developer plan has unlimited seats but short data retention. ConfigCat's free tier allows only ten flags. The usual failure is not the cap itself but a free plan lacking audit logs, single sign-on or approval workflows that a growing team needs.
Can feature flag software also run A/B tests?
Most can, but depth varies a lot. GrowthBook, Statsig and Optimizely were built around experimentation, with statistical methods such as sequential testing and variance reduction, and they connect flag exposures to metrics. Flagsmith and DevCycle offer A/B testing as an add-on to flagging. If experiments are a core activity, check whether results are computed in your own data warehouse or inside the vendor's system, because that decides who owns the history. If you only need occasional split tests, flag-first tools are simpler and cheaper.
What is OpenFeature and does it matter?
OpenFeature is a vendor-neutral specification for how applications call a flag service. Your code talks to the OpenFeature API, and a provider plugin translates calls to whichever product you use. It matters because it makes switching vendors a matter of changing a provider rather than rewriting every flag check. Flagsmith, DevCycle and FeatBit state support for it on their own sites. It does not move your targeting rules between systems, so it reduces code lock-in but not configuration lock-in.
How do we keep old flags from piling up?
Treat flags as short-lived by default. Tag each flag as a release flag, an experiment, an operational kill switch or a permanent entitlement, and give release flags an expiry date at creation. Use whatever your product offers to surface stale flags; Unleash lists technical debt tracking and stale flag detection among its features. Add a lightweight rule to the pull request template: removing the flag code is part of finishing the feature. Without a deletion habit, every product in this list turns into a graveyard within a year.
Which feature flag tools can be self-hosted?
From the products we read, Flagsmith (BSD-3-Clause license), Unleash (AGPL-3.0), GrowthBook (open core, mostly MIT) and FeatBit (a free open-source foundation tier) can be run on your own servers at no license cost. Flagsmith also sells private cloud, and ConfigCat describes an on-premise option. LaunchDarkly, Statsig, DevCycle and Optimizely are described on their sites as hosted services. Self-hosting trades a vendor bill for operational work, so budget for upgrades, backups and someone on call when the flag service has a bad night.
Does it matter whether the vendor is European?
It matters when your customers or regulators ask where flag and user data is processed. Flagsmith is based in London and lists eight hosting regions; Unleash is a Norwegian company that also has a US entity; ConfigCat is Hungarian and states GDPR compliance and ISO 27001:2022 certification. US vendors are not excluded, and several publish regional options, but you should ask where evaluation context (user IDs, attributes) is stored. Self-hosting removes the question for the products that allow it.
Can product managers use these tools without engineers?
Yes, for toggling and targeting, which is much of the point of the category. Role-based permissions and approval workflows let a product manager change a rollout percentage while an engineer approves production changes. The catch is seat pricing: Unleash charges per seat and GrowthBook's Pro plan is per seat, so inviting a dozen non-engineers changes the bill, whereas ConfigCat states unlimited seats on all plans. Decide up front who may change production flags, because a casual toggle in the wrong environment is a common source of incidents.
