Vendor comparison

Split.io alternative: Split.io pricing, Split feature flags, and the Harness FME migration

Split is now sold as Harness Feature Management and Experimentation, and the standalone price card is gone. This page documents exactly what changed, how FME licensing is measured, and what six other flag platforms charge, with every number checked against the vendor's own pricing page in July 2026.

No card required

The short answer

Split.io no longer exists as a product you can buy on its own. Harness acquired Split Software on 11 June 2024 and now sells it as Feature Management and Experimentation (FME). The technology is intact: SDKs, the evaluation engine and the back-end infrastructure all continue to run unchanged. The commercial path is what closed. As of July 2026, split.io/pricing returns a 301 redirect to harness.io/pricing, where FME appears only inside the Enterprise plan behind a Contact Sales button, and licensing is metered on two axes at once, Monthly Tracked Keys and named user seats. If you were on Split for flags and rollouts, the flat-priced platforms below are cheaper and quicker to buy. If you were on Split for its statistical experimentation engine, your realistic alternatives are Statsig or GrowthBook, not a plain flag tool.

Last updated July 2026 · redirects and prices verified against vendor pages on 24 July 2026

01 · What actually changed

The Split.io buying funnel now points at Harness sales

Most write-ups describe the acquisition and stop there, which leaves you guessing whether Split is still a real option. It is checkable. Request each URL and follow the redirects. Here is what happens today.

Split.io URL	Status	Where it lands
split.io/pricing	301	harness.io/pricing (no FME price published)
split.io/product/feature-flags	301	harness.io/products/feature-management-experimentation
split.io/contact-us	301	harness.io/company/contact-sales
app.split.io	200	Still live for existing customers, console moving to app.harness.io
split.io (homepage)	200	Still served, no self-serve purchase path

The redirect on the product page carries utm_content=wildcard, which tells you it is a path-wide rule rather than three hand-written exceptions. That is the practical signal: this is a wind-down of the standalone brand, not a temporary marketing test. For what the surviving Harness modules actually include, see our breakdown of Harness feature flags.

02 · How FME bills

Two meters running at the same time

Most flag vendors pick one thing to charge for. FME charges for two, and you have to stay under both. Harness documents Monthly Tracked Keys as "the number of unique traffic type identifiers (e.g. anonymous) evaluated by FME SDKs in a given month", with the example that 500,000 unique end users in September counts as 500,000 MTKs.

Seats are counted separately, as the named users who log in to manage flags, experiments, environments or integrations. So a small team serving a large consumer audience pays on traffic, and a large engineering org serving a small B2B audience pays on headcount. Model both before you ask for a quote.

The trap worth naming. MTKs count unique identifiers, not paying customers. If you evaluate flags for logged-out visitors, every anonymous session key you generate is billable traffic. Check how your SDK assigns keys to anonymous users before you estimate anything.

typescript · one wrapper, any vendor live

// The realistic migration shape: wrap the vendor once, then swap it.
// Almost nobody regrets this file. Plenty regret not having it.

import { toggled } from '@toggled/sdk';

export async function isEnabled(flag: string, user: User): Promise<boolean> {
  return toggled.isEnabled(flag, {
    userId: user.id,
    attributes: {
      plan: user.plan,
      country: user.country,
      // Split called this a "traffic type"; most vendors call it a key.
      // Normalising it here is what keeps the swap to one file.
    },
  });
}

// Everywhere else in the codebase stays vendor-agnostic:
if (await isEnabled('checkout-v2', user)) { ... }

If you never wrapped the vendor SDK, that is the first thing to fix, and it is worth doing even if you stay. The full sequence is in our guide to feature flag migration.

03 · The alternatives

Split.io alternatives with prices you can actually read

Every figure below came off the vendor's own pricing page on 24 July 2026. This category revises pricing often, so treat any table older than a quarter as fiction, including this one after October.

Platform	Published entry price	What it meters	Best fit for a Split refugee
Harness FME (was Split)	None published, Enterprise only	Monthly Tracked Keys and named seats	Teams already inside a Harness contract, where FME is an add-on rather than a new vendor
LaunchDarkly	$10 per service connection, plus $8.33 per 1k client MAU	Service connections and client-side MAU	Enterprises that want the deepest targeting rules and the widest SDK coverage, and can model a two-part meter
Statsig	$150/mo Pro with 5M events, then $0.05 per 1k	Metered events (exposures, logs, ingested metrics)	Anyone who was on Split specifically for experimentation and needs a real statistics engine
ConfigCat	$0 free, $110 Pro, $325 Smart, $900 Enterprise	Flags, environments and products. Seats unlimited on every tier	Teams that want a flat invoice and never want to count users again
Flagsmith	$0 free, $40 Start-Up, $250 Scale-Up	API requests, plus capped team members per tier	Teams that need self-hosting or data residency without an enterprise negotiation
Unleash	$75 per seat per month, 5 seat minimum self-hosted	Seats, with 53M API requests included	Small teams that want open source lineage, with the caveat below
Toggled	Flat, unlimited seats and flags	Nothing per-user, nothing per-request	Teams tired of a meter deciding how they instrument their own product

Two caveats that matter more than any price. Unleash has deprecated its open source edition, with long-term support that began 10 December 2025 and end of life set for 31 December 2026, which we covered in the Unleash open source end of life. And LaunchDarkly restructured its tiers in 2026 around service connections and AI runs, so older comparisons quoting a per-seat LaunchDarkly price are describing a plan that no longer exists. Current numbers are on our LaunchDarkly pricing page.

04 · The honest version

Three reasons to stay on Harness FME

We sell a competing product, so treat this section as the one to read most carefully. There are real cases where leaving is the wrong call.

Reason 1

You already buy Harness

If your CI, CD or cloud cost modules are already on a Harness contract, FME lands as a line item on a renewal you were signing anyway. That is usually faster and cheaper than adding a seventh vendor to procurement, and the sales motion is one conversation instead of two.

Reason 2

You depend on Split experimentation

Split built a genuine statistics engine: metric definitions, significance calculations, guardrail metrics. Most flag platforms do not have an equivalent and are not pretending to. Swapping to a plain flag tool means rebuilding that analysis somewhere else, and that project is measured in months.

Reason 3

Nothing is technically forcing you

The SDKs still work. Harness states the back end is unchanged and that no application code changes are required. There is no announced shutdown date for the evaluation service. Migrating purely because a brand changed hands is how teams spend a quarter to end up where they started.

The case for leaving is narrower and mostly commercial. You cannot get a price without a sales call, the meter is tied to your traffic volume rather than your team, and the product roadmap now answers to a platform strategy rather than to flag customers. If your use of Split was flags, targeting and rollouts, you are paying an experimentation-platform price for a feature toggle system.

05 · If you do move

What a Split migration actually costs you

The SDK swap is not the expensive part, which is why migration estimates are usually wrong in the same direction. These five items are where the time goes.

Step 1

Inventory the flags that are still alive

Export the flag list and mark each one as permanent, in-flight or dead. In most codebases a third of the list is dead and does not need migrating at all. Doing this first is the single biggest reduction in scope available to you, and it is useful work even if you stay.

Step 2

Translate targeting, do not copy it

Segment definitions and rule precedence differ between vendors, and this is where silent behavior changes hide. Rewrite each rule deliberately and write a test that asserts the same user gets the same answer. Assume nothing maps one to one, because it does not.

Step 3

Normalise the user key

Split calls it a traffic type, others call it a key or a context. If your identifiers do not line up exactly, percentage buckets reshuffle and a user who was in the treatment group lands in control. That looks like a bug in the feature, not in the migration.

Step 4

Run both in parallel

Evaluate old and new for the same request and log disagreements for a week before cutting over. This is the step teams skip under deadline pressure and the one that catches the rule translation errors from step two while they are still cheap.

Step 5

Trace the downstream data

Exposure events feed dashboards, warehouse tables and sometimes billing logic. Changing vendors changes the shape of those events. Find every report that reads them before the old stream stops, not after somebody notices a chart went flat.

Step 6

Set removal dates on arrival

A migration is the one moment when every flag has an owner looking at it. Give each surviving flag an expiry date as you port it, or you will carry the same debt into the new tool. Our feature flag best practices guide covers the hygiene rules worth adopting on day one.

If you are comparing more than one escape route, the full field is in best feature flag tools, and the other common displacement case is covered in LaunchDarkly alternative. Teams that want the swap to stay cheap next time should read our take on whether OpenFeature is worth adopting.

06 · FAQ

Questions people actually search

What happened to Split.io?

Harness acquired Split Software on 11 June 2024 and folded the product into the Harness platform as Feature Management and Experimentation, or FME. The engine and the SDKs continue to run. What changed is the commercial surface: as of July 2026 split.io/pricing, split.io/product/feature-flags and split.io/contact-us all redirect to harness.io, and the management console moved from app.split.io to app.harness.io.

Is Split.io still available?

Yes, as Harness FME rather than as a standalone product. Existing integrations keep working, Split back-end infrastructure is unchanged, and the SDKs need no code modifications. But you can no longer buy Split by itself, and there is no published price for FME. New purchases go through a Harness contract and a sales conversation.

How much does Split.io cost now?

There is no published number. Split used to advertise a self-serve tier and a per-seat paid plan. Today split.io/pricing performs a 301 redirect to the Harness pricing page, where FME appears only inside the Enterprise plan behind a Contact Sales button. Licensing is measured on two axes at once, Monthly Tracked Keys and named user seats, so your cost depends on both traffic and headcount.

What is an MTK in Harness FME?

Harness documents Monthly Tracked Keys as "the number of unique traffic type identifiers (e.g. anonymous) evaluated by FME SDKs in a given month." Their own example: if your app evaluates flags for 500,000 unique end users in September, that is 500,000 MTKs. It is a user-shaped meter, so anonymous traffic and logged-out page views count against it the same as paying customers.

What is the best Split.io alternative?

It depends which half of Split you were using. If you mainly needed flags and rollouts, ConfigCat, Flagsmith and Toggled all publish flat prices and unlimited seats. If you needed the statistical experimentation engine, Statsig and GrowthBook are the closest equivalents. If you need self-hosting for data residency, Flagsmith and GrowthBook both ship a source-available core.

Do I have to change my code to move off Split?

The SDK swap is the small part. Flag evaluation calls look nearly identical across vendors, and a wrapper around your own isEnabled function usually contains the change to one file. The expensive parts are re-creating targeting rules, which rarely map one to one, and rebuilding whatever dashboards read your exposure events. Budget for those two, not for the import.

Is Harness FME the same as Harness Feature Flags?

No, and this trips people up. Harness shipped its own Feature Flags module before the acquisition, and FME is the acquired Split product. They are separate modules with separate SDKs and separate limits. New experimentation work is going into FME. If you are evaluating today, confirm which module a quote actually covers before you sign.

Can I still use Split SDKs after migrating to Harness?

Yes. Harness states that developers do not need to modify application code for feature flags or events, and that the Split Proxy, Synchronizer and Evaluator require no updates. Back-end integrations with S3, Segment and mParticle also carry over. The work sits in the admin layer: new SSO setup, a new permissions model, and five Admin API endpoints that changed.

Early access · launching soon

A price you can read without a sales call.

Unlimited seats, unlimited flags, no per-user meter and no per-request meter. Join the early-access list, no card required.

No card required · your data stays yours

Keep exploring