The best LaunchDarkly alternatives, compared honestly
LaunchDarkly is still the most complete feature management platform on the market, and it did something most vendors never do: it removed per-seat pricing entirely. Seats are now free and unlimited on every tier. The catch is what replaced them. The Foundation plan bills $10 per service connection per month and $8.33 per 1,000 client-side monthly active users, so the invoice tracks your infrastructure and your traffic instead of your headcount. This page compares seven alternatives on the thing that actually decides the migration: what each one puts on the meter.
The best LaunchDarkly alternative depends on which meter is hurting:
- Open source without the AGPL problem → GrowthBook: MIT core, $40 per seat on Cloud Pro, or free self-hosted with unlimited traffic.
- Self-hosted with real governance → Unleash: approvals, SSO and SOC2 at $75 per seat, five-seat minimum, AGPL-3.0 core.
- A bill that ignores headcount → ConfigCat: flat $110 per month for Pro, unlimited team members on every plan.
- Flags plus product analytics → PostHog: first 1,000,000 flag requests a month free, then $0.0001 per request.
Narrower cases are in the table below: Flagsmith if you want a BSD-3 core you can fork, Flipt if flags should live in Git, DevCycle if you actually liked LaunchDarkly's model and just want it cheaper.
Why teams look elsewhere
What pushes teams off LaunchDarkly
The product did not get worse. The reasons below are almost entirely about how the bill is computed, and two of them are things teams discover only after the first invoice.
Seats are free, servers are not
The Foundation plan charges $10 per service connection per month. A service connection is one instance of one server-side SDK connected to one environment. Every application instance across production and pre-production counts, so a staging cluster is a line on the invoice.
MAU counts keys, not people
LaunchDarkly bills $8.33 per 1,000 client-side MAU, and the docs are explicit that usage is the count of unique context keys. Change a key and it counts again. Apps that hand anonymous visitors a fresh key per session inflate this meter by a multiple, not a margin.
The bill scales with success
Under per-seat pricing, cost grew when you hired. Now it grows when your app gets popular or when autoscaling adds instances. That is defensible as infrastructure pricing, but it makes the annual number much harder to forecast at budget time.
The free tier is a demo
Developer is genuinely free forever, but it stops at 5 service connections and 1,000 client-side MAU a month. That is enough to evaluate and not enough to run a small production app, so the step from free to paid arrives early.
A third meter arrived
AI runs are now metered too: 5,000 included, then $5 per 1,000. It is cheap in isolation, but it is a third independent variable on an invoice that used to be headcount times a rate.
No self-hosted escape hatch
LaunchDarkly is proprietary SaaS with no open-source edition. When a pricing model changes, your only options are to renegotiate or to migrate. Four tools on this shortlist let you run the whole thing yourself if it comes to that.
The shortlist
7 LaunchDarkly alternatives worth evaluating
Feature flagging splits into two jobs that vendors bundle differently: flipping flags safely, and measuring what the flag did. Tools that lead with experimentation want your warehouse; tools that lead with flag delivery want your traffic. This ranking optimises for teams leaving LaunchDarkly over cost or control, not for the longest feature list. Each pick lists one honest strength and one real weakness.
Cloud Pro is $40 per seat per month up to 50 users; the self-hosted open-source edition is free with unlimited traffic and one project. Most of the repository is MIT, which is the permissive licence corporate legal teams actually approve. Experimentation runs against your own warehouse, so the statistics are auditable. Weakness: warehouse-native analysis means you need a warehouse to get the value, and per-seat billing punishes giving read access to the whole company.
Pay-As-You-Go is $75 per seat per month with a five-seat minimum for self-hosting, and includes four-eyes approvals, SAML 2.0 and OIDC single sign-on, SOC2 Type II and 90-day metrics retention. Cloud Enterprise includes 53 million API requests a month, then $5 per million. Weakness: the highest per-seat price here, and the open-source core is AGPL-3.0-or-later, which a lot of enterprise legal policies ban outright for network services.
The core is BSD-3-Clause, which is about as unencumbered as licences get, and self-hosting is a first-class path rather than an afterthought. Start-Up is $40 a month billed yearly for three team members and up to a million requests. Weakness: the free tier stops at 50,000 requests a month, which a chatty client SDK burns through quickly, and RBAC plus SAML sit in a closed-source Enterprise Edition.
The first million flag requests each month are free, then pricing tiers down from $0.0001 per request to $0.00001 above 50 million. The repository is MIT expat apart from the ee directory, and a posthog-foss build strips proprietary code entirely. Flags sit next to analytics, session replay and experiments, so you can see what the flag changed. Weakness: if you only want flags, you are adopting a considerably larger platform than the job requires.
Every plan includes unlimited team members, which makes it the cleanest answer for organisations that hated paying per seat and do not want a usage meter either. Free covers 10 flags and 5 million config downloads; Pro is $110 a month for 100 flags and 25 million downloads; Smart is $325 for unlimited flags. Weakness: on Free and Pro the ceiling is the flag count, not traffic, and there is no open-source edition to fall back on.
Flipt v2 keeps flag state in a Git repository, so a flag change is a commit: it reviews, diffs and rolls back exactly like code, and your flags survive the vendor. All SDKs and client libraries stay MIT. Weakness: v2 is under the Fair Core License, which is source-available and non-compete rather than OSI open source, converting to MIT two years after each release; a Pro licence key also gates some features. Git-native means product managers get no UI to change targeting.
If the LaunchDarkly model suited you and only the price did not, DevCycle is the like-for-like swap: unlimited seats, MAU-based billing, an edge delivery network. Free covers 1,000 client-side MAU; Business is $500 a month billed annually with 100,000 MAU included and $2.50 per additional 1,000. Weakness: it inherits the same problem you may be leaving, since the bill still scales with your users, and the gap between free and $500 has nothing in it.
Two names are deliberately absent. Statsig is an excellent product whose ownership changed twice in under a year: OpenAI announced a $1.1 billion acquisition on 2 September 2025, and on 5 May 2026 Amplitude took on the brand and customer base while the engineering team stayed at OpenAI. Optimizely's chief executive called that arrangement a race car without a driver. If you are migrating to escape instability, that is the wrong direction. Split is also out, having been absorbed into Harness in 2024. And for a team of three shipping one app, a flags table in your own database is still a legitimate answer.
Side by side
LaunchDarkly alternatives compared
Prices as of August 2026, from each vendor's own pricing page. The column that decides most migrations is "Bills by": these seven products meter six different units, so list prices are not directly comparable and a per-seat quote can beat a usage quote by an order of magnitude in either direction. LaunchDarkly's own row is the baseline.
| Tool | Type | Bills by | Free tier | Self-host / licence | Best for |
|---|---|---|---|---|---|
| LaunchDarkly | Full feature management | Service connections ($10/mo) + client-side MAU ($8.33/1k). No seat charge. | 5 connections, 1k MAU | ✗ proprietary SaaS | Large teams, small deployments, enterprise governance |
| GrowthBook | Flags + warehouse experiments | Seats: $40/seat/mo (Cloud Pro, to 50 users) | 3 users cloud; unlimited traffic self-hosted | ✓ MIT core, commercial Enterprise dirs | Most teams leaving on cost or licence grounds |
| Unleash | Full feature management | Seats: $75/seat/mo, 5-seat minimum | OSS self-host only | ✓ AGPL-3.0-or-later | Self-hosted approvals, SSO, audit |
| Flagsmith | Flags + remote config | API requests: $40/mo yearly for 1M | 50k requests/mo, 1 member | ✓ BSD-3-Clause core, closed EE | Teams that want a forkable core |
| PostHog | Analytics suite with flags | Flag requests: $0.0001 each, tiering down | 1,000,000 requests/mo | ✓ MIT expat, except ee dir | Seeing what the flag actually did |
| ConfigCat | Flag delivery | Flat tier: $110/mo Pro, unlimited members | 10 flags, 5M downloads/mo | ✗ proprietary SaaS | Predictable invoices, wide team access |
| Flipt | Git-native flags | Self-hosted free; Pro licence key, price on request | Full v2 server | ✓ Fair Core License, MIT after 2 years | GitOps shops that want flags in review |
| DevCycle | Full feature management | Client-side MAU: $500/mo for 100k, then $2.50/1k | 1,000 MAU, unlimited seats | ✗ proprietary SaaS | A cheaper like-for-like swap |
Numbers that do not fit the cells: LaunchDarkly also meters AI runs at $5 per 1,000 past 5,000 included. Flagsmith's Scale-Up is $250 a month billed yearly, with members beyond five at $50 each. Unleash Cloud Enterprise includes 53 million API requests a month, then $5 per million. ConfigCat runs above Pro at $325 (Smart), $900 (Enterprise) and $4,500 (Dedicated) per month. Pricing and free tiers change often; check each vendor for current terms. Compiled August 2026.
Official pages: LaunchDarkly · GrowthBook · Unleash · Flagsmith · PostHog · ConfigCat · Flipt · DevCycle
A fair call
When LaunchDarkly is still the right choice
Migrating flags is cheaper than migrating most infrastructure, but it is not free: evaluation calls port easily, and targeting rules, segments, experiment history and approval workflows generally get rebuilt by hand. Before you start, check whether the new meter is actually better for your shape of usage. For a good number of organisations it is not.
LaunchDarkly is still right if…
- You have a large organisation and a small deployment. Free unlimited seats is a real discount: 200 people can have access and the bill does not move. Per-seat rivals at $40 to $75 would charge you $8,000 to $15,000 a month for the same access.
- You are server-side only, on a B2B product with modest client-side traffic. The MAU meter barely registers, and you are paying a handful of service connections for the most mature platform in the category.
- You need governance that has already passed an audit: approval workflows, fine-grained roles, full audit trails and the compliance paperwork to match.
- You run experiments. The Developer plan includes 100,000 experimentation MAU a month at no charge, which is a genuinely generous line most competitors do not match.
- The problem is your context keys, not the vendor. If anonymous users get rotating keys, fix the keys first; the bill can drop sharply without changing anything else.
Look elsewhere if…
- You run a high-traffic consumer app: the MAU meter is the whole invoice, and ConfigCat's flat tiers or PostHog's request pricing break the link to user count.
- You run a large permanently-scaled fleet across prod and staging: every instance in every environment is a service connection, and GrowthBook self-hosted has no equivalent meter.
- Your compliance position requires running the control plane yourself: Unleash and Flagsmith are built for it, though check AGPL-3.0 against your policy first.
- You want flag changes to go through code review: Flipt stores state in Git, so a flag is a pull request.
- You cannot forecast the annual number, and finance needs one: a flat or per-seat plan is worth real money in predictability even if the headline rate looks worse.
Common questions
Common questions about LaunchDarkly alternatives
What is the best LaunchDarkly alternative in 2026?
There is no single winner, because the tools do not bill for the same thing. GrowthBook suits most teams leaving LaunchDarkly: $40 per seat per month on Cloud Pro, or free self-hosted under MIT with unlimited traffic and one project. Unleash is the pick when you need a self-hosted platform with approvals and SSO, at $75 per seat per month with a five-seat minimum. ConfigCat is the pick when you want a flat bill that ignores headcount entirely, at $110 per month for Pro. Match the tool to the meter that is hurting you, not to the feature list.
Does LaunchDarkly still charge per seat?
No. Seats are free and unlimited on every LaunchDarkly tier, including Enterprise. The Foundation plan meters two other things instead: $10 per service connection per month and $8.33 per 1,000 client-side monthly active users per month, plus $5 per 1,000 AI runs past the included 5,000. This is a genuine improvement if you have a large team and a small deployment, and a genuine problem if you run many server instances or have a high-traffic consumer app. The bill now tracks infrastructure and traffic rather than headcount.
What counts as a service connection in LaunchDarkly?
LaunchDarkly defines a service connection as one instance of one server-side SDK connected to one environment, measured in minutes, assuming 43,800 minutes per month. Only server-side SDKs count; client-side SDKs feed the MAU meter instead. Every application instance across production and pre-production environments counts. Because the meter is time-weighted, two instances that each run for half a month are billed as one service connection, so short-lived autoscaled pods cost a fraction of a full one. A permanently scaled fleet across prod and staging, however, bills for every instance in every environment.
Why is my LaunchDarkly MAU bill higher than my user count?
Because LaunchDarkly counts unique context keys, not people. The docs state that usage is based on the total number of unique context keys evaluated in a month, and that changing an existing key makes it count as an additional MAU. If your app assigns anonymous or rotating keys, a single logged-out visitor across five sessions can register as five monthly active users. Fixing this is a code change rather than a plan change: give anonymous contexts a stable key, stored client-side, before you conclude the platform is too expensive.
Is there a truly open-source LaunchDarkly alternative?
Yes, but read the licences carefully, because they differ more than the marketing suggests. GrowthBook is MIT for most of the repository, with several directories under a separate commercial Enterprise licence. PostHog is MIT expat except its ee directory, and it publishes a posthog-foss repository with all proprietary code stripped. Flagsmith's core is BSD-3-Clause, with RBAC and SAML in a closed Enterprise Edition. Unleash is AGPL-3.0-or-later, not Apache-2.0 as many comparison posts still claim. Flipt v2 uses the Fair Core License, which is source-available and non-compete, converting to MIT two years after each release.
Can I migrate off LaunchDarkly without a rewrite?
Flag evaluation calls port more easily than teams expect, because most SDKs expose the same shape: a client, a flag key, a user or context object, and a default value. What does not port is everything around the flags. Targeting rules, segments, experiment history, audit logs and approval workflows are vendor-specific and generally have to be rebuilt by hand. The practical path is to run both systems in parallel behind a thin internal wrapper, move one service at a time, and keep LaunchDarkly on its free Developer tier during the overlap rather than cancelling on day one.
Should I use Statsig as a LaunchDarkly alternative?
Statsig is a strong product, but its ownership has changed twice in under a year, so it is hard to recommend as a stable landing spot in mid-2026. OpenAI announced its acquisition of Statsig on 2 September 2025 for $1.1 billion. On 5 May 2026 Amplitude announced it was taking on the Statsig brand and customer base, while the original engineering team stayed at OpenAI. Optimizely's chief executive publicly described the arrangement as a race car without a driver. If you are migrating specifically to escape platform instability, that is worth weighing.