Split
Harness · Ranked #5 of 7 in Feature Flag APIs
Mature feature-management-and-experimentation platform (now Harness FME) with strong impression-data-driven metrics and 15+ open-source SDKs.
Feature management + experimentation (Harness FME)
Overview
Split (split.io) is a feature management and experimentation platform that pairs feature flags with built-in, statistically-rigorous A/B testing. Its differentiator versus pure flag tools has always been "measurement at the flag", when you roll a flag out, Split automatically attributes the rollout to downstream business and operational metrics (conversion, latency, error rates) and runs significance testing, so a single flag doubles as an experiment. SDKs evaluate flags locally against an in-memory cache that is kept in sync via a Fastly-backed CDN and a streaming architecture, so treatment evaluation is effectively local (sub-millisecond) and flag-definition changes propagate "in milliseconds." This makes it a strong fit for product and growth engineering teams who want to tie release decisions to data, rather than teams that only need simple on/off toggles.
The most important fact about Split today is corporate: in 2024 Split was acquired by Harness and rebranded as Harness Feature Management & Experimentation (FME). split.io now redirects into the Harness estate (pricing, status, and docs all live on harness.io / developer.harness.io), and FME is positioned as a module within the broader Harness software-delivery platform. The standalone Split brand, pricing transparency, and roadmap are being absorbed accordingly, which review sites flag as a source of uncertainty for buyers who chose Split as an independent best-of-breed tool.
Split's recurring weaknesses are cost and opacity. Historical public pricing started around $33/seat/month with metered "monthly traffic keys" (≈MTUs) that require custom quotes to scale, advanced capabilities (data export, SSO, audit logs, dedicated support) gated to Enterprise, and Growth/Enterprise pricing hidden behind sales. Reviewers also note weaker native CI/CD integration than LaunchDarkly and that the experimentation depth is overkill for teams who just want flags. It is well-regarded where it counts (G2 ~4.6/5), but it is a premium, experimentation-first product, not a lightweight or cheap one.
How this score is derived
The APIbenchmarks Index is a weighted sum of four dimensions, each scored on an absolute 0–100 reference scale. See the methodology for every mapping.
| Dimension | Score | Weight | Contribution |
|---|---|---|---|
| Documentation & DXExtensive, well-structured docs and a large per-language SDK reference now hosted on the Harness Developer Hub and the legacy Split Help Center, covering SDKs, synchronizer, proxy, and evaluator. | 80 | 30% | 24.0 |
| ReliabilityBuilt on AWS with feature-flag definitions served via Fastly's CDN and a public status page (status.split.io, now redirecting to status.harness.io); SDKs evaluate locally so apps stay up even during control-plane incidents. | 84 | 25% | 21.0 |
| Ecosystem & SDKsBroad official SDK coverage across client and server languages plus integrations, but reviewers consistently call out weaker native CI/CD pipeline integration than LaunchDarkly, with some connectors needing webhooks/Zapier. | 82 | 25% | 20.5 |
| AccessibilityA free tier exists for evaluation but is tightly capped (limited flags/seats) and most pricing now requires a sales conversation following the Harness acquisition, lowering self-serve accessibility. | 78 | 20% | 15.6 |
| APIbenchmarks Index (ABI) | 81.1 | ||
Table 1. Derivation of the ABI for Split. Contribution = score × weight; the index is their sum.
At a glance
- Vendor
- Harness
- Pricing model
- Per-seat (usage-based at scale)
- Free tier
- Free Developer tier: up to 10 seats, 50 feature flags
- Official SDKs
- 12 languages
Pricing
| Free / Developer | $0 | Evaluation tier; tightly capped (reported limits around 10 feature flags / up to 10 seats). Core flag functionality to get hands-on. |
| Team (Starter) | from $33 / seat / month | Unlocks unlimited flags, advanced targeting and deeper metric integrations; historically capped at ~50,000 monthly traffic keys (≈MTUs) with extra keys priced via custom quote. |
| Business / Growth | Custom (contact sales) | Adds scale and additional capabilities; pricing not published. |
| Enterprise | Custom (contact sales) | Full functionality including data export, SSO, audit logs and dedicated support. Now bundled into the Harness platform's enterprise tier. |
Key features
- •Feature flags with percentage rollouts and progressive delivery
- •Built-in A/B testing and experimentation with guardrail metrics and automated significance detection
- •Automated impact measurement (attributes flag rollouts to conversion, latency, error rates)
- •Customer segmentation ("splits") on user attributes
- •Instant kill switch for disabling features
- •Local in-memory SDK evaluation with Fastly CDN delivery and streaming updates
- •Audit logging for compliance
- •REST Admin API for programmatic flag control
- •Split Evaluator (Dockerized REST service) for languages without a native SDK
- •Self-hosted infra options: Split Synchronizer and Split Proxy
Official SDKs
Strengths & trade-offs
- +Experimentation is first-class: rolling out a flag automatically attributes it to business/operational metrics with statistical significance testing
- +Local, in-memory SDK evaluation backed by Fastly CDN + streaming means sub-millisecond treatment evaluation and millisecond flag propagation
- +Strong, granular customer segmentation/targeting ("splits") on user attributes
- +Broad official SDK coverage across both client-side and server-side languages, plus a REST Split Evaluator for unsupported languages
- +Clean, intuitive console UI with comprehensive audit logging and a kill-switch for instant rollback
- +Highly rated by users (G2 ~4.6/5 across ~300 reviews)
- –Premium pricing, generally more expensive than LaunchDarkly, with metered "traffic keys"/MTUs that make scaling costs hard to predict
- –Pricing is opaque: Growth/Enterprise require contacting sales, and most advanced features (data export, SSO, audit logs) are gated to Enterprise
- –Weaker native CI/CD pipeline integration than competitors; some connections need custom webhooks or Zapier
- –Experimentation/analytics depth can be overkill (added complexity) for teams who just want simple on/off flags
- –Free tier is too limited (≈10 flags) to realistically evaluate at team scale
- –Harness acquisition + rebrand to FME creates uncertainty about the standalone product's roadmap and independence
What developers say
G2 4.6/5 (~300 reviews)
Users praise Split's data-driven experimentation, fast local evaluation and clean UI, while criticism centers on premium/opaque pricing, weak CI/CD integration, and uncertainty after the Harness acquisition.
“Split is rated 4.6/5 on G2 from around 300 reviews; many users prefer it because setting up and managing feature flags in one place is easier and the UI isn't confusing.”
Key figures
| Flag definition global delivery latency | < 200 ms via dual-layer CDN (Fastly) | Split.io product/architecture material ↗ |
| Flag change / update propagation | milliseconds (streaming-enabled SDK) | Split Help Center / Harness Developer Hub ↗ |
| Starter (Team) plan price | from $33 / seat / month | TrustRadius / comparison pricing ↗ |
| Starter plan included traffic keys (≈MTUs) | ~50,000 monthly traffic keys (overage = custom quote) | Split pricing analysis (comparison) ↗ |
| Service status / uptime tracking | Public 90-day status & uptime history (status.split.io → status.harness.io) | Split by Harness status page ↗ |
Compare Split head to head
Sources
- https://www.split.io/
- https://help.split.io/hc/en-us/articles/360033557092-SDK-Overview
- https://developer.harness.io/docs/feature-management-experimentation/sdks-and-infrastructure/
- https://www.harness.io/blog/split-io
- https://www.g2.com/products/split/reviews
- https://www.trustradius.com/products/split/pricing
- https://flagshark.com/blog/launchdarkly-vs-split-vs-unleash-comparison-2026/
- https://status.split.io/uptime
- https://www.split.io/blog/how-i-evaluated-my-feature-flags-in-milliseconds-with-splits-new-integration-for-vercel/
Figures last verified 2026-06-27. Spotted an error? corrections@apibenchmarks.com