Best Feature Flag Tools in 2026

LaunchDarkly was founded in 2014 in Oakland, California, by Edith Harbaugh and John Kodumal, and was one of the first companies to sell feature flagging as a stand-alone product. The company reports more than 5,500 customers, says roughly a quarter of the Fortune 500 use it, and states that about 45 trillion flags are evaluated on its platform each day; it lists offices in Oakland and London.

The platform has expanded beyond flags into an observability and release suite: the pricing page meters service connections, client-side monthly active users (MAU), AI runs under AgentControl, session replays, captured errors, trace spans and log lines. Its Developer plan is free with unlimited seats, five service connections, 1,000 client-side MAU, one project and three environments. The Foundation plan is pay-as-you-go with unlimited projects and environments, charging US$10 per service connection and US$8.33 per 1,000 client-side MAU on yearly billing (US$12 and US$10 monthly); the published example of 10 connections and 2,000 MAU costs US$58.33 per month on yearly billing. Enterprise is custom-priced. LaunchDarkly removed per-seat pricing, but its many usage meters make bills for large consumer apps with high client-side MAU harder to predict, and experimentation and governance depth sits mainly in higher tiers.

Statsig was started in February 2021 and is based in Bellevue, Washington; its chief executive is Vijaye Raji. It built a combined platform for feature gates, dynamic configs, experimentation, product analytics and session replay. On 2 September 2025 it announced a definitive agreement to join OpenAI. Its website now states that "Statsig is part of the Amplitude family", and the footer names Statsig, LLC. Buyers should confirm the current contracting entity and roadmap, as the product has changed hands twice in under a year.

Statsig's pricing is unusually generous for flags. The free Developer tier includes 2 million metered events, 50,000 session replays and one-year analytics retention per month, with unlimited seats and unlimited gate and config checks. Statsig does not by default charge for exposures from 0% or 100% roll-outs; partial roll-outs with Metric Lifts enabled are metered. Pro costs US$150 per month with 5 million events included and US$0.05 per 1,000 events thereafter; Enterprise is contract-priced. If the free event limit is exceeded, targeting continues to work but metric computations stop. The strength is the integration of flags with statistically rigorous experiments; teams that only need simple toggles may find the analytics surface larger than required.

Harness Feature Management & Experimentation (FME) is the former Split product. Harness, a CI/CD platform vendor led by Jyoti Bansal, announced on 29 May 2024 that it would acquire Split Software, with financial terms undisclosed, and said Split would be folded into the Harness brand. FME covers feature flags with progressive rollouts and kill switches, versioned and schema-validated configs including AI configs for swapping prompts and models, A/B and multivariate experimentation with sequential testing and guardrail metrics, a warehouse-native experimentation mode, release monitoring with automated rollback triggers, and governance through role-based access, policies, approvals and audit trails.

Harness also offers a Feature Flag Cleanup Agent that finds stale flags and drafts pull requests, and its AI assistant explains experiment results. The product page lists a free plan for individual developers and small teams; on Harness's main pricing page, FME appears as a module of the Enterprise plan, priced by quotation. Tight coupling with Harness pipelines is the main advantage for existing Harness customers, while teams on other CI/CD tools gain less from it and face opaque pricing above the free tier.

Optimizely Feature Experimentation is the server-side and full-stack flagging and testing product from Optimizely, a digital experience platform vendor best known for web A/B testing and content management. It supports targeted delivery by user, region or custom attributes with permissions and approvals, progressive rollouts, a dashboard kill switch, dynamic configuration that lets non-engineers change copy, limits and pricing tiers at runtime, and flag governance showing each flag's owner, status and environment, with mutual exclusion to keep experiments clean.

SDKs cover backend, mobile and edge environments with sub-millisecond local evaluation, results can be tied to business metrics and the customer's data stack, and an MCP server lets AI assistants manage flags and archive stale ones. Optimizely does not publish prices; its plans page asks buyers to request pricing tailored to their business. The product suits organisations that already run Optimizely for web experimentation or content, where shared metrics and audiences are an advantage. Its limitations are the absence of a published price list or verified free tier and an enterprise sales process that is heavier than self-serve tools.

Unleash is an open-source feature management platform that can be self-hosted or used as a hosted service. Unleash Open Source is free, self-hosted and has no seat limit, but is restricted to one project, two environments and 5,000 flags per instance, and excludes SSO, SCIM, granular role-based access, change-request approvals and a support SLA. Self-hosting keeps flag configuration and user context inside the customer's own environment, a common reason for choosing it in regulated sectors.

The commercial product is sold in two ways. Enterprise Pay-As-You-Go costs US$75 per seat per month, billed monthly by credit card without a contract, available hosted or self-hosted (with a five-seat minimum for self-hosted); the cloud version includes 53 million API requests per month, then US$5 per million, a 99.9% uptime commitment and 90 days of feature metrics. Enterprise annual contracts are quoted per customer, support cloud, self-hosted or hybrid deployment, and offer a 99.99% uptime commitment and add-ons such as a hosted private instance and multi-region Enterprise Edge. A 14-day trial requires no card. The seat-based price is predictable but grows with large engineering teams, and experimentation features are lighter than in Statsig or Optimizely.

PostHog is an open-source product analytics suite whose modules include feature flags alongside analytics, session replay, experiments and surveys. Feature flags are billed by the number of API requests made to evaluate flags at the /flags endpoint, not by the $feature_flag_called events that record exposures. The first million requests each month are free; beyond that, published rates fall from US$0.0001 per request between 1 and 2 million to US$0.00001 per request above 50 million. PostHog states that more than 90% of companies use it for free, requires no card to start and lets customers set billing limits.

Server-side local evaluation reduces costs: each definitions poll costs ten requests when definitions changed and one when they did not, so one server instance polling every 30 seconds uses about 86,400 requests per month, while SDKs without ETag support use about 864,000. The advantage of PostHog is that flags, experiments and the analytics used to judge them share one event pipeline. The limitations are that its flag governance (approvals, environments) is lighter than LaunchDarkly or Unleash Enterprise, and client-side heavy apps can generate many billable requests.

GrowthBook is an open-source feature flagging and experimentation platform designed to analyse experiments against data already in the customer's warehouse rather than shipping events to a vendor. It is offered both as a cloud service and as self-hosted software. The Cloud Starter plan is free for up to three users and one project with unlimited flags, experiments and traffic, and no card is required; Pro costs US$40 per seat per month for up to 30 users and three projects and adds multi-arm bandits, safe rollouts and an AI visual editor; Enterprise is custom-priced.

The self-hosted Open Source plan is free with unlimited users, one project and unlimited flags and experiments, and a self-hosted Enterprise tier adds SSO and SCIM, custom environments and dedicated support. Cloud overages are published: US$10 per additional million CDN requests, US$1 per additional GB of CDN bandwidth and US$0.03 per 1,000 events in the managed warehouse after the first 2 million. The warehouse-native design suits analytics-mature teams, but it assumes a data warehouse and SQL-literate users, and the free tiers are limited to one project.

Flagsmith is an open-source feature flag and remote configuration service available as a hosted cloud, private cloud or self-hosted deployment. Its pricing is published and request-based. The Free plan includes about 50,000 requests per month, 100,000 experiment users and one seat, with unlimited flags, environments and segments. Start-Up costs US$40 per month billed yearly (US$45 monthly) for about 1 million requests, 500,000 experiment users and three seats, adding A/B and multivariate testing and scheduled flags.

Scale-Up costs US$250 per month yearly (US$300 monthly) for 5 million or more requests, 2 million or more experiment users and five seats, expandable to 20 at US$50 per seat per month on yearly billing, and adds SSO/SAML, governance features and priority support. Enterprise, with 20 or more seats and an SLA, is where on-premises and own-cloud self-hosting are offered. Overages are charged per request, for example US$7 per 100,000 requests on Start-Up after a grace period. Flagsmith offers discounts for open-source, not-for-profit and charity projects. Its strengths are deployment flexibility and low entry pricing; its experimentation analytics are less developed than in experimentation-first tools.

ConfigCat is a hosted feature flag and configuration service that charges by config JSON downloads and network traffic rather than by seats or end users. Every plan includes unlimited seats. The Forever Free plan includes 5 million config downloads (about two per second) and 20 GB of traffic per month, but limits accounts to 10 flags, two environments, two products, two segments, four targeting rules and four percentage options per flag, one webhook and seven days of audit-log retention. ConfigCat states that the free plan has the same feature set as the paid plans, only with lower limits.

Paid plans are published in several currencies: Pro at US$110 per month for 25 million downloads and 100 GB, Smart at US$325 for 250 million and 1 TB, Enterprise at US$900 for 1 billion and 4 TB, and Dedicated at US$4,500 for 6 billion or more and 24 TB, with annual billing saving up to 10%. Because billing follows config downloads rather than end users, costs do not rise with the size of the user base. The model is easy to forecast for server workloads; the free tier's ten-flag cap makes it more of an evaluation tier, and experimentation analytics are not a focus.

DevCycle is a feature flag platform originally operated by Taplytics Inc.; its pricing page states that DevCycle is now part of Dynatrace and that references to Taplytics now mean the Dynatrace subsidiary providing the service. DevCycle emphasises the OpenFeature standard, listing OpenFeature support across all of its SDKs, which reduces lock-in because application code calls a vendor-neutral API.

The Free plan includes unlimited seats and flags, up to 1,000 client-side MAU, 10,000 cloud config requests, 100,000 server config requests and 5,000 events per month, plus A/B testing and debugging tools. Business is listed at US$500 per month billed annually (US$625 monthly base) with 100,000 MAU, then US$2.50 per 1,000, 1 million cloud and 10 million server config requests and 500,000 events, then US$0.50 per 10,000. Enterprise adds custom SSO/SAML and an uptime SLA. MAU is defined as a unique user with at least one client-side SDK initialisation in a month. The Dynatrace ownership is attractive for Dynatrace customers, but the jump from free to US$500 per month leaves little middle ground for small teams.

Flipt is an open-source, self-hosted feature flag server that stores flag state in Git. Changes made in its user interface become Git commits, so every flag change is reviewable through the same workflow as code, and it ships as a single binary with no external dependencies. The Open Source edition is free with unlimited flags, Git-native workflows, real-time updates, REST and gRPC APIs and community support.

Flipt Pro costs US$200 per month with a 14-day trial and adds up to ten instances, integrations with GitHub, GitLab, Bitbucket and Azure DevOps, pull requests created from the UI, GPG-signed commits, integrated secrets management and email support with a 24-hour response time. Enterprise is custom-priced and adds unlimited instances, distribution rights for customer and air-gapped installations, invoice or purchase-order billing and security-review assistance; annual billing saves 20%. Flipt is well suited to GitOps-oriented infrastructure teams and air-gapped environments, but it has no hosted SaaS option comparable to LaunchDarkly, and analytics and experimentation are outside its scope.

AWS AppConfig, a capability of AWS Systems Manager, manages feature flags and free-form dynamic configuration for applications running on AWS. Configuration profiles of either type are validated before deployment, rolled out gradually through deployment strategies, and automatically rolled back if an Amazon CloudWatch alarm fires. Configuration can be sourced from the AppConfig hosted store, Secrets Manager, Parameter Store or Amazon S3, and applications retrieve it through the AppConfig Agent, which caches the deployed configuration locally. Access is controlled through IAM, with KMS encryption and CloudTrail auditing.

AppConfig experimentation adds A/B tests and gradual rollouts using production traffic, works on EC2, Lambda, ECS and EKS through the agent, and can export assignment data to Redshift, Snowflake or Databricks for analysis. Pricing is pay-as-you-go: US$0.0000002 per configuration request via API calls and US$0.0008 per configuration received. AppConfig is inexpensive and well integrated for AWS-hosted services, but it lacks the targeting user interface and non-engineer workflows of dedicated flag vendors and is not intended for applications outside AWS.

Firebase Remote Config is a Google cloud service that changes an app's or server's behaviour without releasing an update: developers define in-app defaults and override them from the Firebase console or backend APIs. Conditions target users by app version, language, Google Analytics audience, imported segment or custom signals; rollouts release features gradually using parameters as feature flags and compare results against an automatically provisioned control group using Crashlytics and Analytics data; and real-time Remote Config lets apps receive newly published values without polling.

A project can hold 3,000 parameters per template type, keep up to 300 template versions and run up to 24 A/B experiments and rollouts combined. On Firebase's pricing page, Remote Config is listed as up to 100,000 requests per day per project on the no-cost Spark plan and no-cost up to that level on the Blaze plan, with charges for usage above it starting on 1 September 2026; A/B Testing is listed as no-cost. Remote Config is the default choice for Firebase mobile apps, but its tight coupling to Google Analytics and Firebase, limited governance and per-project limits make it less suitable as an organisation-wide flag system.