← Blog

Split vs LaunchDarkly: feature management, two ways

Split (now sold as Harness Feature Management and Experimentation) and LaunchDarkly are both commercial, dashboard-hosted feature-flag platforms, but they bet on different things. Split treats every flag as a measurement, built for teams that want experiment stats tied directly to the rollout. LaunchDarkly bets on breadth: the widest SDK coverage and the most integrations of any flag platform on the market.

Neither keeps a flag in your repo. Both store flag state and targeting rules in a hosted service you configure through a web dashboard, and both ship an SDK that streams rule changes to your app without a deploy. The split vs LaunchDarkly question usually comes down to which bet matters more for your team: measurement depth, or platform reach.

What changed since Harness acquired Split in 2024 matters here too, and so does where an open-source, git-native tool like dif fits if the dashboard model itself, in either platform, is the problem.

Key Takeaways

  • Split and LaunchDarkly are both closed-source, dashboard-hosted platforms. Neither keeps flag state in your repo.
  • Split treats every flag as a measurement, with mature impression-level stats. LaunchDarkly treats experimentation as a separate, added layer on top of flagging.
  • LaunchDarkly moved off per-seat pricing in 2026: its Foundation plan now runs about $10 per service connection per month plus $8.33 per 1,000 client-side monthly active users. Split’s Teams tier is still priced around $33 per seat per month, with custom contracts for enterprise.
  • LaunchDarkly wins on SDK breadth and integrations; Split’s edge is experiment measurement built into the flag itself.
  • Since the 2024 Harness acquisition, Split’s roadmap and pricing sit inside a larger delivery platform rather than standing alone.

Where flags live: two dashboards, not your repo

In both tools, a flag is a row in a hosted service. You create it and set its targeting and rollout percentage through a web UI. The SDK in your app polls or streams the current state, so a change in the dashboard takes effect without a deploy.

That gives both platforms something a git-native tool does not: an instant toggle. It also means the flag and the code that checks it live in two separate systems. The dashboard has to be kept in sync with the code by hand, because nothing ties a split.getTreatment("new-checkout") call or a LaunchDarkly variation() call back to a single file a reviewer can read in a pull request.

For teams specifically looking to move that reconciliation into git, dif’s open-source LaunchDarkly alternative covers the repo-native model in depth.

Split vs LaunchDarkly pricing

LaunchDarkly moved off per-seat billing in 2026. The free Developer tier has unlimited seats. The paid Foundation plan meters usage instead: roughly $10 per service connection per month, plus $8.33 per 1,000 client-side monthly active users. Cost now tracks how much of the product you use, not how many people touch the dashboard.

Split has a free Developer tier for small usage. Above it, the Teams plan has been priced around $33 per seat per month, and since the Harness acquisition most published pricing above that tier has given way to custom enterprise contracts. Per-seat pricing means a 50-person team costs roughly five times what a 10-person team costs, independent of how many flags or experiments either runs.

Neither pricing model is wrong. A usage-metered platform like LaunchDarkly’s Foundation plan scales with traffic; a per-seat platform like Split’s Teams tier scales with headcount. Which one is cheaper depends on your ratio of engineers to traffic.

Experimentation: a measurement vs a bolt-on layer

Split’s defining idea is that a feature flag and an experiment are the same object. Every rollout can be a treatment, with impression data captured automatically and a stats engine built to analyze it. If you are already flagging a change, turning it into a rigorous experiment is a configuration change, not a new system.

LaunchDarkly’s flagging product is the core, and experimentation is a separate add-on with its own setup. It works well, but it was not the platform’s original bet, and a LaunchDarkly experiment takes more setup than a Split treatment does.

For teams where the primary question is “how good is the experiment stats engine,” Split wins this comparison. dif’s Split comparison goes deeper on what Split’s measurement model buys you and what it costs.

Platform breadth: SDKs, integrations, enterprise reach

LaunchDarkly has the broadest SDK coverage of any flag platform: every major language, mobile, edge runtimes, and a long list of integrations into CI/CD, observability, and incident tools. For a large engineering org running a mix of languages and platforms, that breadth is the reason to pick LaunchDarkly over almost anything else, including Split.

Split’s SDK coverage is solid but narrower, and its integration list is smaller. Since the Harness acquisition, the strongest integration story is with Harness’s own CI/CD platform: teams already running Harness for delivery get flags in the same place they manage pipelines.

Ownership: Harness vs an independent roadmap

Harness acquired Split in 2024, and the product is now sold as Harness Feature Management and Experimentation. The flagging and experimentation core is still Split’s, but the roadmap, pricing, and support now sit inside a larger platform company with its own priorities.

LaunchDarkly has stayed an independent, standalone platform through its pricing changes. For a team deciding between the two, that is a real difference in how much the tool’s future depends on decisions made elsewhere.

What neither gives your coding agent

A coding agent working in your repo cannot log into the Split or LaunchDarkly dashboard. It has no way to see which flags exist or what they are set to, because that state lives behind a login screen the agent never opens. In either platform, an agent can delete a branch behind a flag it does not know is live.

dif closes that gap. dif build emits a context.json, a machine-readable summary of every active flag and each surface’s latest learning, and a coding agent like Claude Code or Cursor reads it on session start. Neither Split nor LaunchDarkly has an equivalent, because the model that makes an instant dashboard toggle possible is the same model that keeps the state out of the repo an agent reads.

Who should choose which

Choose LaunchDarkly if you run many languages and platforms, need the broadest integration list, and want the most mature, battle-tested dashboard-based platform on the market.

Choose Split (Harness FME) if experiment measurement tied directly to the flag matters more than platform breadth, or you already run Harness for delivery and want flags in the same place.

Choose an open-source, git-native tool like dif if the dashboard model itself, in either platform, is the problem: you want flags reviewed in a pull request, versioned in git, and readable by a coding agent, and you are willing to trade an instant dashboard toggle for a deploy-based rollback.

FAQ

Is Split or LaunchDarkly better? Neither is better outright. LaunchDarkly has broader SDK and integration coverage; Split has a stronger built-in experimentation model, since every flag can be a measured treatment. The better one depends on whether platform breadth or experiment depth matters more to your team.

Is Split the same company as LaunchDarkly? No. They were always separate companies. Harness acquired Split in 2024 and now sells it as Harness Feature Management and Experimentation, while LaunchDarkly has remained an independent company.

Is LaunchDarkly cheaper than Split? It depends on your team’s shape. LaunchDarkly’s Foundation plan metres usage (about $10 per service connection plus $8.33 per 1,000 MAU), so it scales with traffic. Split’s Teams plan is priced around $33 per seat per month, so it scales with headcount. A small team with high traffic may find LaunchDarkly cheaper; a large team with low traffic may find Split cheaper.

Is there an open-source alternative to both? Yes. dif is an open-source, MIT-licensed tool that keeps flags and experiments as Markdown files in your repo instead of a hosted dashboard. It has no per-seat fee, and the CLI and SDK are free.

Can I migrate from Split or LaunchDarkly to dif? You re-author each active flag as a Markdown file and replace the vendor’s SDK call with dif’s at the render site. Since dif flags are plain files, you write the ones worth keeping and drop the rest, with no export to reconcile.

Getting started

Split and LaunchDarkly both keep feature management in a hosted dashboard, with the tradeoffs that come with that model: an instant toggle, and state your coding agent cannot see. If you want flags as files in your repo instead, reviewed in pull requests and readable by an agent, dif’s feature flags page covers the model. Install the CLI and scaffold a project:

npm install -g @dif.sh/cli
dif init

dif init writes the dif/ directory, the generated client, and the managed agent blocks. dif Cloud pricing covers the optional hosted analytics layer, at a flat $50/month for 1M events with unlimited seats.

Pricing and product facts are drawn from each vendor’s own site as of 2026, including LaunchDarkly’s pricing page, the Harness Feature Management and Experimentation product page, and the Harness acquisition announcement. Verify current pricing with each vendor, since tiers and enterprise contracts change.