Feature Flags 2026: LaunchDarkly Stopped Charging for Seats
LaunchDarkly now bills per server-side SDK connection and per client MAU with unlimited seats, which turns the old advice about moving evaluation server-side upside down. Unleash relicensed to AGPL in May and its open-source Edge dies on 31 December.
Three things changed in this category over the past few months, and none of them show up on a comparison grid.
LaunchDarkly deleted per-seat pricing. Its Developer plan now says “Unlimited seats. No per-seat pricing,” and the paid Foundation tier says “No platform or seat fees.” What it bills instead is $10 per service connection per month, where a service connection is one server-side SDK connected to one environment, plus $8.33 per thousand client-side monthly active users on annual billing. Five service connections are included, then the meter starts.
Read that meter twice if you have ever taken the standard advice about LaunchDarkly bills, which was to move flag evaluation out of the browser and into your backend so the MAU meter stops spinning. That advice is now half a bill. Server-side evaluation is exactly what a service connection is, and every environment you evaluate in counts separately. Ten services across dev, staging, and production is thirty connections, $250 a month after the included five, before a single end user shows up.
The second change is that Unleash relicensed. The third is that its open-source edge proxy has a date on it. Both of those are worth their own section.
Unleash moved to AGPL, and you can see exactly when
The LICENSE file in the Unleash repository was Apache 2.0 for eleven years. On 26 May 2026 a commit titled “task: Move to AGPL-v3” replaced it with the GNU Affero General Public License v3.
You can check which releases that caught. Tag v7.6.5, published 29 June 2026, still carries Apache 2.0. Tag v8.0.0, published 9 June 2026, carries AGPL-3.0, as does v8.1.0 from 5 August. So the line is the major version, not the date: 7.x is Apache, 8.x is AGPL, and the 7.6 branch kept getting Apache-licensed patches after the switch landed on main.
AGPL on a server you run internally is not the problem people assume it is. If you deploy Unleash for your own employees and do not offer it to third parties, the network clause does not force you to publish anything. It matters if you were planning to embed a modified Unleash in something you ship, or if your company has a blanket policy against copyleft in the deployment path, which plenty do. The pull request that made the change notes that the Docker image becomes AGPL too.
Unleash Edge has 130 days left
The bigger issue for anyone self-hosting Unleash is in the unleash-edge README, under a warning callout: “The open-source version of Unleash Edge is in long-term maintenance mode, with end-of-life scheduled for December 31, 2026.”
Edge is the read-replica proxy that lets one Unleash instance serve thousands of connected SDKs. The main Unleash README recommends it in the security and performance section, as the way to “scale with Unleash Edge independently of the Unleash server.” So the open-source project points you at a component that the same organization has scheduled for end of life, and the replacement is Enterprise Edge.
Stack that against what open-source Unleash already gives you. Two environments. No projects. No role-based access control, no single sign-on, no advanced segmentation, all listed in the README’s own “looking for more features” section. The free tier was already narrow, and the part that made it scale is leaving.
I would not start a new self-hosted Unleash deployment right now. If you are already on one and using Edge, the date to put in the calendar is 31 December 2026, which is 130 days from when I checked this.
The price sheet
| Platform | Free tier | Entry paid | Meters on | Self-host license |
|---|---|---|---|---|
| LaunchDarkly | 5 connections, 1k MAU, 1 project | $10/connection + $8.33/1k MAU | Connections + client MAU | none offered |
| Flagsmith | 50k requests, 1 member | $40/mo yearly, 1M requests | Members + requests | BSD-3-Clause |
| Unleash | 2 environments, no projects | $75/seat/mo, 53M requests | Seats | AGPL-3.0 from 8.0.0 |
| PostHog | 1M requests | usage only, no seat fee | Requests | MIT with paid directories |
| GrowthBook | 3 users, 1 project | $40/seat/mo cloud | Seats (cloud only) | MIT with paid directories |
LaunchDarkly’s own page quotes both $8.33 per 1,000 client MAU billed yearly and $10 per 1,000 in the monthly comparison table, so the number you get depends on the contract. Flagsmith lists $40 a month billed yearly and $45 monthly for Start-Up, $250 and $300 for Scale-Up, with extra team members at $50 or $60 each. PostHog’s request bands run $0.0001 for the first million past the free million, then $0.000045, $0.000025, and $0.00001 above fifty million.
The same team, five bills
Ten engineers, a hundred thousand client-side monthly actives, twenty server-side SDK and environment pairs, about five million flag evaluations a month.
LaunchDarkly Foundation: fifteen billable connections at $10 is $150, plus 100 × $8.33 for MAU is $833. Total $983, seats free.
Flagsmith Scale-Up on annual billing: $250 for the plan and five million requests, plus five members beyond the included five at $50 each. Total $500.
Unleash pay-as-you-go: ten seats at $75 is $750, with the 53 million request allowance nowhere near touched. Self-hosted has a five-seat minimum on the same price.
GrowthBook Cloud Pro: ten seats at $40 is $400. Self-hosted open source is $0.
PostHog: one million free, one million at $0.0001, three million at $0.000045. That is $235, and nobody is counting your engineers.
The spread is four to one between the cheapest and the most expensive, which is much tighter than the same exercise a year ago. Per-seat pricing at this size is no longer the outlier it was, because the vendor that made seats expensive stopped selling them.
What “free self-hosted” leaves out
All three open-source options run from a Docker image in about ten minutes and none of them are complete.
Unleash open source is two environments and no projects, covered above. GrowthBook’s self-hosted comparison marks single sign-on, sequential testing, CUPED variance reduction, exportable audit logs, and custom roles as Enterprise, which trims most of the statistical sophistication people cite as the reason to pick GrowthBook. Flagsmith is the most permissive of the three at the file level, BSD-3-Clause across the majority of the platform, with governance features behind an Enterprise license key.
The operational cost is the same for all of them and it is not the container. A flag service sits on the request path of everything that reads a flag. If it is down and your SDKs have no local cache and no bootstrapped defaults, you do not degrade, you fail. Every SDK worth using supports both; if you self-host, configuring both is not optional. Budget Postgres with a real backup story, a cache, two replicas, and a couple of hours a month forever.
OpenFeature moves the SDK, not the rules
OpenFeature is the CNCF vendor-neutral flag SDK specification, currently at spec v0.9.0. It works, and it does remove the part of a migration where you rewrite call sites in every service.
What it does not carry is everything that makes the migration expensive: targeting rules, segments, experiment definitions, audit history, approval workflows. Those live in each vendor’s data model and there is no interchange format for them. Adopt OpenFeature, because a cheaper SDK swap is still cheaper. Do not let it convince you that you can change vendors over a weekend.
Picking one
If you want flags today with no purchase order, PostHog. No seat charge, a million evaluations free, and the whole team can be in it.
If you already pay for a warehouse and want experiment analysis on top of it, GrowthBook, cloud if you need SSO and the better statistics, self-hosted if you do not.
If you are on LaunchDarkly and the bill hurts, do the arithmetic before the migration, and do it in the new units. Count service connections, which means SDK instances times environments, not services. That number is often larger than teams expect and it is the one you can actually reduce, by consolidating environments rather than by rewriting where flags are evaluated.
And if you are self-hosting Unleash, the license change is probably fine and the Edge date probably is not. Check which one applies to you before December.
Prices, license files, and the Edge end-of-life notice were all read from vendor pricing pages and GitHub repositories on 23 August 2026. The metering models in this category changed more in the last year than the feature sets did, so treat any table older than a quarter as a starting point for your own arithmetic. The same pattern of a free tier quietly narrowing showed up in HCP Terraform earlier this year.