5 Split Alternatives Now That Split Is Harness FME (2026)

Split mattered for one reason - it was one of only two flag platforms that took experimentation seriously, pairing feature flags with statistically rigorous A/B analysis. Then Harness bought it. The deal completed on 11 June 2024, and Split was rebranded Feature Management and Experimentation, a module inside the wider Harness developer platform.

The engine is fine. The problem is everything around it. Standalone Split pricing is gone - split.io/pricing now 301-redirects to harness.io/pricing. FME is sold as a contact-sales module under the Harness Enterprise plan, on the historical MAU-based model carried forward, with no published per-tier numbers. For anyone who just wanted Split’s flags and experiments as a standalone product, that path is closed. You’re buying into, or migrating into, Harness now.

For existing customers the technical migration was handled well. The back-end is unchanged, so no application code, SDK, Proxy, Synchronizer or Evaluator changes were required. The friction is elsewhere - SSO must be reconfigured as a new Harness connection, five admin API endpoints changed so automation scripts break, and there’s a real repricing risk as the standalone deal turns into a platform-bundle deal at renewal.

So this isn’t a “Split is bad” list. The engine is good. It’s a list for the specific reasons people are leaving the Split-to-Harness transition anyway. I’ve matched each pick to the reason you’re looking.

The short version

Tool	Best for	Starting price	Billing model	Self-host
Harness FME	Experimentation inside a managed platform	Contact sales	MAU-based, part of Harness bundle	No
Statsig	Experimentation-first, free flag checks	$0	Analytics events, flags free	No
GrowthBook	Open-source warehouse-native experiments	$0 self-hosted	Per seat (cloud)	Yes (MIT)
LaunchDarkly	Deepest pure flag tooling	Usage-based	$10/connection + $8.33/1k MAU	No

If you left because the standalone product is gone: Statsig

The most common reason to leave the Harness transition is that you wanted a focused experimentation-and-flags product, not a line item in a DevOps suite. Statsig is exactly that. It’s experimentation-first, built by ex-Facebook engineers to mirror Facebook’s internal stack, and the A/B engine is the strongest in this set - a full statistical engine with sequential testing and CUPED variance reduction.

Its billing is the inverse of an MAU model. Flag and config checks are unlimited and free on every tier, including the free one. Statsig monetizes analytics and experiment events instead. So the flag half of what you had in Split costs nothing, and you pay on the analytics you send. Pro is a flat $150 per month, which is a real number you can budget - rare in this category.

The honest gotcha: OpenAI acquired Statsig on 2 September 2025 for a reported ~$1.1 billion, per third-party reporting, and founder Vijaye Raji became OpenAI’s CTO of Applications. The product still ships independently, but adopting it today is a multi-year bet on a roadmap that now sits inside OpenAI. If you’re leaving Split to escape acquisition churn, note that Statsig has its own ownership story - it’s just a different owner. Full detail on the Statsig page.

If you want experimentation you can self-host: GrowthBook

Split’s experimentation was proprietary and hosted, and FME keeps it that way. If part of what you want now is to own the stack - keep your event data on your own infrastructure - GrowthBook is the pick. It’s warehouse-native, meaning it queries the BigQuery, Snowflake or Databricks warehouse you already run rather than ingesting a copy of your events. The stats engine is the real thing: CUPED, sequential testing, Bayesian analysis, multi-arm bandits, SRM checks.

In ambition it’s the closest open-source peer to what Split was going for. The self-hosted MIT edition runs unlimited users for $0, Cloud Pro is $40 per seat per month for up to 50 users, and there’s a free Cloud Starter for 3. It’s a Y Combinator company with a reported $22.6M Series A backed by Khosla and Nexus, per third-party sourcing.

The gotcha is dependency and difficulty. GrowthBook needs a properly instrumented data warehouse to deliver its value, and the statistics have a genuine learning curve - reviews cite an engineer-first UI and docs that are thorough but hard to navigate. If you have data maturity, GrowthBook is the best-value serious experimentation platform here. If you don’t, you’ll get a fraction of the value.

If you want the deepest pure flag tooling: LaunchDarkly

Some teams used Split mostly for the flags and lightly for experiments. If that’s you, and you want the deepest flag platform available, LaunchDarkly is it - advanced targeting, around 38 SDKs, and guarded releases that automatically roll back on a bad metric, which is genuinely best-in-class.

Be clear-eyed about the trade, because it’s a cost trade. LaunchDarkly bills $10 per service connection per month plus $8.33 per 1,000 client-side MAU, charged on your highest-volume context kind - so 500,000 devices outweigh 300,000 users even if you only target users. Per Vendr’s third-party contract data, the median deal runs around $72,000 a year, ranging from $19,500 to $165,700. You’re leaving one enterprise-priced product for another; the difference is that LaunchDarkly is the flag-depth leader and publishes its meters, where FME is a hidden platform quote.

The gotcha: the renewal. Buyers report proposed renewals at roughly double their current cost on legacy plans, and add-ons stack - at 10M MAU, add-ons alone can exceed $33,700 a month. Model your context volume before you commit. Full breakdown on the LaunchDarkly page.

So which one?

• You want a focused experimentation-and-flags product with free flag checks - Statsig, with the OpenAI-ownership asterisk priced in.
• You want to own the experimentation stack and keep event data in-house - GrowthBook, if you have a warehouse and someone who understands the stats.
• You want the deepest flag tooling and can forecast usage-based cost - LaunchDarkly, and count your highest-volume context kind first.
• The honest case for staying: if you’re already in or open to the Harness ecosystem and want experimentation-grade feature management inside a managed platform, Harness FME is a genuine contender. The engine and SDKs are unchanged, the migration didn’t break anyone’s code, and Harness is well-funded - it raised a $240M Series E in December 2025 led by Goldman Sachs at a $5.5B valuation, per its own announcement. Evaluate it as a Harness commitment, not a point tool, because that’s what it has become.

One note on the numbers. LaunchDarkly’s contract figures are from Vendr’s third-party buyer data, not LaunchDarkly’s pricing page. Every alternative’s price was read from its own tool page, verified against the vendor’s site on 23 July 2026. The right pick is whichever one matches the reason you started reading - a focused product, a self-hostable one, or the flag-depth leader.