The best feature flag services for 2026: an honest comparison

Strip away the dashboards and a feature flag is an if statement with a subscription fee. The market for hosted if statements is now one of the most crowded in developer tools, and almost every comparison page in it was written by a vendor about its competitors — usually aimed at enterprises weighing a LaunchDarkly migration. This one is aimed at the rest of us: small teams, indie developers, and early-stage startups who mostly want to know what’s actually free, what the first invoice looks like, and which product fits how they work.

Full disclosure: I build one of the services on this list: Smpl Flags. Same rules as the cron comparison: I’ve put it where I think it belongs and said where the others are stronger.

The two kinds of “feature flag service”

Flag-first platforms

ConfigCat, DevCycle, Flagsmith, LaunchDarkly, Smpl Flags, and Unleash sell the flag itself: targeting, rollouts, environments, kill switches, and the governance around them. Some offer experimentation too, but it isn’t why you buy them.

Experimentation platforms (flags included)

GrowthBook, Harness FME, PostHog, and Statsig sell measurement. Flags exist to randomize who sees what, and the statistics engine that tells you what happened next is the actual product. If you adopt one of these, you’re adopting the analytics stack that comes with it — and some of them barely charge for the flags at all.

The free-tier truth table

Every vendor page in this category has a pricing table arranged to flatter the house: the unlimited things in big type, the metered things in a tooltip. Here is the un-fudged version — the real free-tier limits and the first paid price, each verified against the vendor’s live pricing page on August 1, 2026.

Service	Free seats	Free usage / mo	Free flags	Free envs	First paid price
ConfigCat	Unlimited	5M config downloads	10	2	$120/mo
DevCycle	Unlimited	1,000 MAU	Unlimited	Unlimited	$625/mo
Flagsmith	1	50K requests	Unlimited	Unlimited	$45/mo
GrowthBook	3	1M CDN requests	Unlimited	4 built-in	$40/seat/mo
Harness FME	Not published	Not published	Not published	Not published	Contact sales
LaunchDarkly	Unlimited	"Limited"	Unlimited	3	$10/conn + $8.33/1K MAU
PostHog	Unlimited	1M flag requests	No stated cap	None	Usage-based
Smpl Flags	1 writer	Unlimited	10	2	$99/mo
Statsig	Unlimited	Flag checks unmetered	No stated cap	3 defaults	$150/mo
Unleash	No free cloud tier — open-source self-hosting instead				$75/seat/mo

Prices are the vendor's published monthly-billing rate; annual discounts are noted in each section. LaunchDarkly publishes only billed-yearly rates. Notes on the odd cells: LaunchDarkly's free tier caps service connections and client-side MAU at levels its pricing page describes only as "Limited" — no numbers. Harness enforces free-plan limits but publishes none of them. PostHog has no environments product (the documented pattern is one project per environment) and its paid pricing is per request past the free million — about $100 a month at 2M requests. Statsig's meter is analytics events (2M/mo free), never flag checks. Smpl Flags allows one console writer free with unlimited read-only users.

The three meters, and what they do to your bill

Feature-flag pricing comes in three basic shapes, and the shape matters more than the sticker price.

Per seat — GrowthBook at $40, Unleash at $75. The bill tracks your team size and ignores your traffic entirely. Predictable, and kindest to high-traffic products built by small teams. It stings later, when you want the whole company reading the dashboard and every login is a line item.

Per MAU — DevCycle and LaunchDarkly. The bill tracks how many people use your product. Nothing about your flag setup has to change for the invoice to double; your user count did that on its own. It’s a tax on success, which is at least the right thing to tax.

Per request — Flagsmith and PostHog meter flag requests; ConfigCat meters config downloads. Here your architecture is the pricing lever, because it decides what a “request” even is. With remote evaluation, every flag check is an API call: one check per second exhausts Flagsmith’s free 50,000 requests in about half a day. With local evaluation, the SDK downloads the ruleset, evaluates in-process, and the meter only sees polling — the same application, orders of magnitude fewer billable calls. Before you compare prices per million, find out which mode your SDKs will actually run in.

And two vendors decline to meter flag usage at all: Statsig doesn’t charge for flag checks on any plan — its meter is analytics events — and Smpl Flags has no usage meter, pricing by managed-flag count instead. Harness meters “monthly tracked keys,” which would slot into the MAU family if it published prices. It doesn’t.

The field at a glance

Service	Type	% rollouts	Experimentation	Self-host	OpenFeature
ConfigCat	Flags-first	✓	✗	Dedicated plan	✓
DevCycle	Flags-first	✓	✓	✗	✓
Flagsmith	Flags-first	✓	Enterprise beta	✓	✓
GrowthBook	Experimentation	✓	✓	✓	Partial
Harness FME	Experimentation	✓	✓	✗	✓
LaunchDarkly	Flags-first	✓	✓	✗	Partial
PostHog	Experimentation	✓	✓	Hobby only	Partial
Smpl Flags	Flags-first	✗	✗	✗	✗
Statsig	Experimentation	✓	✓	✗	Community
Unleash	Flags-first	✓	✗	✓	✓

✓ first-class support  ·  ✗ not offered  ·  partial cells explained in each section

Category 1: flag-first platforms

ConfigCat

ConfigCat has the most readable pricing model in this lineup: every feature ships on every plan — SAML SSO included, on the free tier — seats and MAU are unlimited everywhere, and the only things you pay for are limits. Free means 10 flags in 2 environments with 5 million config downloads a month; the SDKs fetch config from a CDN and evaluate locally, so flag checks themselves are never network calls and never billed. Targeting covers arbitrary custom attributes, segments, and percentage options, with 22 SDKs, OpenFeature providers for 14 platforms, a CLI, and a Terraform provider. What’s missing at any price: built-in experimentation (variation IDs export to your analytics tool), approval workflows, and scheduled changes — the last two listed as in development on their news page. The first paid plan is $120 a month ($110 billed annually). It’s the rare pricing page you can read in one sitting and believe.

DevCycle

DevCycle is the OpenFeature-native pick: its SDKs are built around the standard, with providers across client and server languages. The free tier is generous in most directions — unlimited seats, flags, environments, and projects, with A/B testing included — except the one your product grows in: 1,000 client-side MAUs. The next plan is Business at $625 a month ($500 billed annually) for 100,000 MAUs. Between the two plans there is nothing. Server SDKs do local bucketing with SSE-pushed updates; approvals are Enterprise-only and the audit log starts at Business. One more thing to price in: Dynatrace acquired DevCycle in January 2026, and the product continues standalone “in the near term.” Their words.

Flagsmith

Flagsmith is the open-source pick that also has a proper cloud: a BSD-3-licensed core you can self-host free, or a hosted free tier with 50,000 requests a month — and one seat. Start-Up at $45 a month ($40 annually) gets you 3 seats and a million requests. Server SDKs in nine languages can evaluate locally; client SDKs call the API per fetch, which is what the request meter counts. Change requests (approvals) and audit logs arrive at Scale-Up, $300 a month, and real-time push updates are Enterprise-only. Native experimentation appeared recently as an Enterprise beta; everyone else pipes multivariate results to their own analytics tool. One thing we can’t resolve for you: the pricing page says scheduled flags are included at Start-Up, and the docs say Scale-Up and above. Both pages were live the day we checked. Bring a screenshot.

LaunchDarkly

LaunchDarkly defined this category, and it is still the most capable platform on this page: the deepest targeting, experimentation included at every tier, about 30 SDKs, and rule changes streamed to connected SDKs in real time. Two things to understand before reaching for it. First, pricing is usage-based on two meters — service connections at $10 a month each and client-side MAU at $8.33 per thousand, billed yearly — and every environment your servers connect to counts as a connection, so the bill compounds quietly. Second, the free Developer tier includes unlimited seats and flags but caps connections and MAU at levels the pricing page now describes only as “Limited.” The numbers used to be published. Approvals and flag scheduling live at Enterprise. If you’re a platform team with governance requirements, this is the reference implementation, and worth it. If you’re three people with a launch date, you’d be paying reference-implementation prices.

Smpl Flags

Smpl Flags has no usage meter at all: API requests, connected servers, and read-only users are unlimited on every plan, including free. Plans are priced by managed-flag count — free covers 10 flags in 2 environments with one console writer, and $99 a month raises that to 50 flags, 3 environments, and unlimited writers. Flags are typed (boolean, string, numeric, JSON), optionally constrained to a declared value set, and targeted with JSON Logic rules over arbitrary context attributes; SDKs in six languages evaluate locally with WebSocket-pushed updates, and there’s a CLI and a Terraform provider. The distinctive part is that it’s code-first: flags referenced by running code surface in the console automatically, no registration step. Now the honest part. There are no percentage rollouts — the only service in this lineup without them — no built-in experimentation, no OpenFeature provider, and no self-hosting. If staged percentage rollouts are your daily workflow, pick something else today.

Unleash

Unleash is the self-hosting standard-bearer. Backend SDKs evaluate locally by default — user data never leaves your process — and the open-source core runs on your own hardware for free, though deliberately shaped: one project, two fixed environments, basic roles, and an AGPL license as of v8. Change requests (a genuinely good four-eyes approval flow), SSO, and custom environments belong to the paid tier. There is no free cloud tier; hosted Unleash starts at $75 per seat per month with the full enterprise feature set and 53 million requests included. Experimentation analytics are external by design — impression data flows to whatever analytics tool you already have. One date for self-hosters: the open-source Unleash Edge reaches end of life on December 31, 2026, with the Enterprise edition as its designated successor. If OSS Edge sits in front of your frontend SDKs, that’s a migration to plan this year.

Category 2: experimentation platforms

GrowthBook

GrowthBook is experimentation-first done the warehouse-native way: your event data stays in your warehouse, GrowthBook connects read-only, runs the statistics, and shows you the results. The honest prerequisite is that you need a data warehouse and the discipline to log exposure events — or their cloud-only managed warehouse, which removes both requirements. The flag side is solid: attribute targeting with MongoDB-style conditions, percentage rollouts, and local SDK evaluation from a cached payload. The free cloud tier covers 3 seats, 1 project, and a million CDN requests a month; Pro is $40 per seat. Approvals and audit logs are Enterprise, and the open-source core self-hosts free with the commercial features license-gated. Best when the measurement engine is the reason you showed up — if “data warehouse” isn’t a thing your team has, start elsewhere.

Harness FME (Split)

Harness FME is Split, acquired by Harness in 2024, and the Split engine remains the real thing: a mature stats engine with a warehouse-native option, environment-level approval flows, per-environment permissions, audit logs with webhooks, and local-evaluation SDKs with sub-second streaming updates. OpenFeature providers cover nine platforms. The pricing is where the small-team story ends: there’s a free plan and a contact-sales Enterprise plan, and the free plan’s limits — flags, environments, tracked keys, seats — are enforced as “service limits” you can see in your account but that appear on no public page. In a comparison where every other number came from a pricing page, Harness is the vendor without one. A fine choice if your company already runs Harness; otherwise, the first feature you’ll use is the sales call.

PostHog

PostHog bundles flags into its analytics suite, and if you already live there, they’re right where you’d want them: person and cohort targeting, percentage rollouts, experiments wired into the same event data, unlimited seats, and a million flag requests free every month. Past that it’s usage-based — about $100 a month at 2 million requests — though server-side local evaluation changes the math dramatically, since polling bills a flat ten requests per fetch and the per-user checks themselves are free. The trade-offs are structural: there are no environments — the documented pattern for dev/staging/prod is three separate projects and a copy tool — approvals sit in the Enterprise package, and self-hosting is an officially unsupported hobby deployment. Best when PostHog is already your analytics home and the flags are one more reason not to leave.

Statsig

Statsig doesn’t charge for flag checks. On any plan. The meter is analytics events — 2 million a month free, with unlimited seats — so a small team using flags without heavy experimentation has the most free headroom in this lineup, indefinitely. The experimentation engine is the actual product and it’s a serious one: CUPED, holdouts, switchback tests, warehouse-native deployment at Enterprise. Server SDKs evaluate locally in memory; approvals arrive at Pro, $150 a month. The asterisk is corporate: OpenAI acquired Statsig in September 2025, and in May 2026 the brand and customers moved to Amplitude, which has publicly committed to maintaining and developing the platform. The product hasn’t visibly missed a beat. It has also changed hands twice inside a year. Price that risk yourself.

Which should you pick?

If you want free, simple, and honest limits, use ConfigCat: unlimited seats, every feature on every plan, and a meter that local evaluation barely touches. If you want the most free headroom, Statsig is the strongest free tier in this lineup for a flags-only team — go in with eyes open about the ownership story. And if your whole company already lives in PostHog, the flags are one login away and effectively free at small scale.

If self-hosting is the requirement, Unleash and Flagsmith are the serious options, with GrowthBook close behind when experimentation is the point. Just read the fine print on what the open-source cores withhold — approvals and SSO live behind the paid tiers on all three.

If you want to measure, not just toggle: GrowthBook if you have a data warehouse, Statsig or PostHog if you don’t, Harness FME if your organization already runs Harness and doesn’t mind pricing by conversation.

LaunchDarkly is more capable than everything else here. Pick it if you’re a platform team that needs the governance — approvals, scheduling, guardrails — and will actually use the depth. Skip it if you’re a small team, not because it wouldn’t work, but because you’d be renting a container ship to cross a river.

And if you want flags without a usage meter — no MAU count, no request quota, unlimited servers — with typed flags that register themselves from code, Smpl Flags deserves serious consideration. It won’t run your experiments or your percentage rollouts; it will run your flags, predictably, for a price that doesn’t move when your traffic does.

Self-hosting and standards: a short aside

Three things worth knowing that aren’t full contestants. OpenFeature is the CNCF-incubated standard API for feature flags: write your evaluation calls against it and every service in this lineup except Smpl Flags has a provider — though LaunchDarkly’s, GrowthBook’s, PostHog’s, and Statsig’s coverage is partial or community-maintained, so check your language before betting on it. If vendor lock-in worries you, this is the hedge. Flipt is the self-hosted single-binary option whose v2 stores flags in Git — changes are commits — with the caveat that the v2 server moved to a fair-source license. FeatBit is the MIT-licensed, .NET-stack alternative, self-hosted via Docker or Kubernetes. None of these is a managed service. All of them are part of why the managed services can’t charge whatever they like.

One more thing