Datadog

Datadog feature flags: pricing, feature flag tracking, and what the GA product really requires

Most of what is written about Datadog and feature flags is out of date, because the situation changed twice: Datadog acquired Eppo in May 2025 and shipped a native flag-serving product in February 2026. This is the current picture: which of the three Datadog things you actually mean, what each costs, and the infrastructure the native product quietly requires.

No card required

The short answer

"Datadog feature flags" now means three different products, and conflating them is the most common mistake. Datadog Feature Flags is a native flag-serving platform that went generally available on February 3, 2026: it creates flags, evaluates variants and serves them through OpenFeature-based SDKs. Feature Flag Tracking is the older RUM integration that ingests flag values from other vendors (LaunchDarkly, Statsig, Split, Flagsmith and more) purely so you can correlate variants with errors and session replays. It cannot serve a flag. Datadog Experiments is the Eppo product, acquired in May 2025, which runs warehouse-native statistics. Pricing for the native product is free below 1 million monthly flag configuration requests and $55 per month billed annually from 1M to 10M. The catch worth knowing before you plan a migration: server-side evaluation requires Datadog Agent 7.55+, APM tracing and Remote Configuration, so you cannot use Datadog flags on a backend without adopting Datadog APM first.

Last updated July 2026 · verified against docs.datadoghq.com and datadoghq.com/pricing

01 · Disentangle it first

Three Datadog products, one confusing name

Almost every comparison article still describes only the second one, because it was the only one that existed until early 2026. Work out which you mean before you read anything else.

GA February 3, 2026

Datadog Feature Flags

A real flag-serving platform. You create flags in Datadog, the SDKs receive the configuration, and your application evaluates variants. Targeting rules, traffic splitting, environments, approvals, client and server SDKs, all built on the OpenFeature specification. This is the product that competes with LaunchDarkly.

The older RUM integration

Feature Flag Tracking

A reporting sink, not a flag service. You call addFeatureFlagEvaluation to tell Datadog which variant your existing vendor already served, and Datadog attaches it to the RUM session so you can filter errors and watch replays by variant. If this is all you use, you still need a flag vendor.

Formerly Eppo, acquired May 6, 2025

Datadog Experiments

The experimentation and statistics layer. It connects to your warehouse (Snowflake, BigQuery, Redshift, Databricks) and measures how changes moved business metrics. Assignment can come from a Datadog feature flag or from your own randomization.

02 · Wire up tracking

One line per evaluation, and one trap in the flag name

If you are keeping your current flag vendor and only want variant-level observability, Feature Flag Tracking is genuinely cheap to adopt. Enable the experimental feature at RUM init, then report each decision.

Datadog documents integrations for Amplitude, ConfigCat, DevCycle, Eppo, Flagsmith, GrowthBook, Kameleoon, LaunchDarkly, Split and Statsig, plus custom setups. Most work the same way: a callback in the vendor's SDK forwards the decision into RUM.

The trap: Feature Flag Tracking does not support a long list of special characters in flag names, including the dot, the colon and the hyphen. Dot and dash separated keys like checkout.new-flow are the near-universal convention at every other vendor, so this bites on day one of a migration and it bites silently.

Minimum SDK versions: Browser 4.25.0, iOS 1.16.0, Android 1.18.0, Flutter 1.3.2, React Native 1.7.0.

browser · @datadog/browser-rum · flag tracking live

// Feature Flag Tracking: reporting ANOTHER vendor's decision into RUM.
// This does NOT serve flags. It tells Datadog what already happened.
import { datadogRum } from '@datadog/browser-rum';

datadogRum.init({
  applicationId: process.env.DD_APPLICATION_ID,
  clientToken: process.env.DD_CLIENT_TOKEN,
  site: 'datadoghq.com',

  // Still required after ~3 years GA. Without it, nothing is recorded.
  enableExperimentalFeatures: ['feature_flags'],
});

// Call this wherever your flag vendor hands you a decision.
datadogRum.addFeatureFlagEvaluation('checkout.new-flow', variant);

// GOTCHA: these characters are NOT supported in tracked flag names:
//   .  :  +  -  =  &&  ||  >  <  !  ( ) { } [ ] ^ " ~ * ? \
// Dot- and dash-separated keys are the norm everywhere else, so
// "checkout.new-flow" above is exactly the shape that breaks.

On mobile the method names mirror this: RUMMonitor.shared().addFeatureFlagEvaluation(name:value:) on iOS and GlobalRumMonitor.get().addFeatureFlagEvaluation(key, value) on Android.

03 · The spec sheet

Everything Datadog documents, in one table

Taken from Datadog's own docs and pricing pages in July 2026. Where a row says "not documented", it means we could not find it published, which is not the same as it not existing.

Dimension	What Datadog documents
Native product status	Datadog Feature Flags, generally available February 3, 2026
Standard	Built on OpenFeature, the vendor-neutral flag API specification
Price, under 1M MFCR	Free. Usage below 1 million monthly flag configuration requests is included at no cost.
Price, 1M to 10M MFCR	$55 per month billed annually, $89 billed monthly, $100 on demand
Price, above 10M MFCR	Contact sales
What an MFCR is	Not documented. Datadog publishes the price of a flag configuration request but no definition of what counts as one.
Client-side evaluation	Local and instantaneous. The SDK uses locally cached data, so no network request occurs at evaluation time.
Server-side prerequisites	Agent 7.55 or later, API key, APM tracing enabled, Remote Configuration enabled, plus DD_EXPERIMENTAL_FLAGGING_PROVIDER_ENABLED=true
Agent polling	Default 60 seconds. Client telemetry flush defaults to 10,000ms.
Evaluation context	Flat primitive values only. Nested objects and arrays cause exposure events to be silently dropped.
Stale flag detection	Three 30-day signals: fully rolled out, no evaluations, or no modification. Can generate a removal PR.
Governance	Five org-level permissions including approvals override, plus per-flag edit restriction by user, service account, role or team
Feature Flag Tracking scope	Reporting only. Correlates variants with RUM Explorer, Error Tracking and Session Replay. Cannot serve a flag.
Tracking integrations	Amplitude, ConfigCat, DevCycle, Eppo, Flagsmith, GrowthBook, Kameleoon, LaunchDarkly, Split, Statsig, plus custom
Tracking flag-name limits	Dot, colon, hyphen, plus and many other special characters are unsupported
Tracking billing	No separate SKU exists, and Datadog publishes no billing statement either way for Feature Flag Tracking. It rides inside RUM sessions you already pay for.
RUM session pricing	RUM Measure $0.15 per 1,000 sessions annually ($0.22 on demand). RUM Investigate $3 ($4.50). Session Replay add-on $2.50 ($3.60). List rates, US region.
RUM session definition	Expires after 15 minutes of inactivity, capped at 4 hours, then a new session starts. Sessions and replays are retained 30 days.
RUM sampling	Retention Quotas (in Preview) cap retained sessions per day, so flag analysis inherits RUM sampling rather than being a census
Migration tooling	The Statsig guide routes through Eppo as an intermediate step. No LaunchDarkly migration guide found.
Flag-count caps, rate limits, retention	Not documented
Serving SLA	Not documented. Notable for a component sitting in your request path.
Site availability	Not supported on certain government cloud sites

One number is missing from that table on purpose. Datadog publishes what a monthly flag configuration request costs but not what one is, and whether an MFCR is a config poll, an evaluation or a client init changes the bill by an order of magnitude. Ask before you sign.

04 · The walls

Four things to settle before you migrate

Wall 1

Server-side flags require the whole Datadog stack

This is the big one. Datadog documents Agent 7.55+, an API key, APM tracing enabled in your application, and Remote Configuration enabled org-wide as prerequisites for server-side evaluation. If you were hoping to try Datadog flags on one backend service without adopting Datadog APM, that is not a supported path. The flag product is a reason to buy the platform, which is the point.

Wall 2

The billing unit is undefined in public docs

Datadog publishes tiers priced per monthly flag configuration request, and no page we could find defines what a flag configuration request is. Config poll, flag evaluation and client initialization are wildly different volumes. Free under 1M is a great headline; it is not a number you can plan against until someone tells you what it counts.

Wall 3

GA, but still behind an experimental env var

Server-side evaluation needs DD_EXPERIMENTAL_FLAGGING_PROVIDER_ENABLED set to true, and the older RUM tracking still requires enableExperimentalFeatures with feature_flags. Both work. Both are also a fair signal about maturity for something you are about to put in the request path.

Wall 4

Migration tooling has not caught up with the launch

Datadog's own Statsig migration guide routes customers to Eppo as an intermediate step before fully migrating to the dedicated Feature Flags product, and we found no LaunchDarkly migration guide at all. If you are on LaunchDarkly today, budget for a hand-rolled migration, plus the flag-name character restrictions if you also want tracking.

The tripwire · silent drops

A nested attribute in your evaluation context discards the exposure event

Datadog documents that evaluation context attributes must be flat primitives: strings, numbers and booleans. Nested objects and arrays are not supported and will cause exposure events to be silently dropped. Nothing errors. Your flag still evaluates, your rollout still looks healthy, and the exposure data underpinning any analysis of that rollout is quietly incomplete. If you are used to passing a whole user object as context, that habit has to change before the first flag ships.

05 · The honest fork

Should you use Datadog for flags? Sometimes clearly yes.

Use Datadog

It is the right call when

Datadog is already your system of record for APM, RUM and logs, and the Agent is everywhere already.
You want rollback driven by telemetry rather than by a human watching two dashboards side by side.
Your volume sits under 1M flag configuration requests, where the native product is genuinely free.
You want experimentation with warehouse-native statistics from the same vendor, which is what Eppo brought.
You are keeping your current flag vendor and only want variant-level correlation. Feature Flag Tracking is one line of code and no new contract.

We should be straight about the money here: at $55 per month for up to 10M requests, Datadog undercuts most of this category including our own planned mid tier. If you are already all-in on Datadog and that band fits you, it is a strong offer and you should take it.

Look elsewhere

Stay with a dedicated vendor when

You need flags on backends without adopting Datadog APM, the Agent and Remote Configuration first.
You are deliberately avoiding observability lock-in, or you are multi-cloud with a heterogeneous estate.
You need mature migration tooling off LaunchDarkly today rather than a hand-rolled cutover.
You need a contractual serving SLA for something sitting in your request path.
You need government cloud support, where the Feature Flags product is not available.
You cannot plan a budget around a unit the vendor has not defined in public.

A standalone feature flags platform stays independent of your observability vendor, which also means a bad day for one is not a bad day for both. See how the wider field compares on best feature flag tools.

06 · FAQ

Questions people actually search

Does Datadog have feature flags?

Yes, and this changed recently. Datadog Feature Flags became generally available on February 3, 2026 as a real flag-serving product with targeting rules, traffic splitting and SDKs. Before that, Datadog only had Feature Flag Tracking, which reports flag values from other vendors into RUM. Both products exist today and they do very different jobs.

What is the difference between Datadog Feature Flags and Feature Flag Tracking?

Feature Flag Tracking is a reporting sink: you tell Datadog which variant your existing vendor already served, and Datadog correlates it with RUM errors and session replays. It cannot serve a flag. Datadog Feature Flags is the native product that actually evaluates and serves flags. If you only use the tracking integration, you still need a flag vendor.

How much do Datadog feature flags cost?

Datadog publishes three tiers for the native product: free below 1 million monthly flag configuration requests, then $55 per month billed annually for 1M to 10M ($89 monthly, $100 on demand), and contact sales above 10M. Datadog does not publish a definition of what counts as a flag configuration request, which makes the real cost hard to model.

Is Datadog feature flag tracking free with RUM?

There is no separate SKU for Feature Flag Tracking, and Datadog publishes no billing statement about it either way, so it effectively rides inside RUM sessions you already pay for. It is not free of consequence though. In the native Feature Flags SDK, enableRumFlagEvaluationTracking defaults to true, and Datadog documents that this can increase RUM-billed event counts.

Why did Datadog acquire Eppo?

Datadog announced the Eppo acquisition on May 6, 2025 to expand into AI product analytics and experimentation. Eppo connected directly to existing data warehouses and ran a statistical engine measuring how product changes affected business metrics. Eppo became Datadog Experiments, and Datadog says its Feature Flags product was created with a head start from that acquisition.

Does Datadog Feature Flags support OpenFeature?

Yes. Datadog documents its Feature Flags product as built on OpenFeature, the vendor-neutral specification for feature flag APIs. That matters for portability: you are coding against a standard interface rather than a proprietary one, so the call sites in your application are not the hard part of any future migration.

What does addFeatureFlagEvaluation do in Datadog RUM?

It is the RUM SDK method that reports a single flag evaluation to Datadog so the value can be attached to the session. You call datadogRum.addFeatureFlagEvaluation(key, value) after your flag vendor returns a decision. It requires Browser RUM 4.25.0 or later and the feature_flags experimental flag enabled at init.

Can you use Datadog feature flags without the Datadog Agent?

Not for server-side evaluation. Datadog documents the server-side prerequisites as Agent 7.55 or later, an API key, APM tracing enabled in your application, and Remote Configuration enabled for your organization. That is a heavy dependency chain: you cannot put Datadog flags in a backend without adopting Datadog APM infrastructure first.

Early access · launching soon

Want flags that do not require an observability platform underneath

Unlimited flags and seats, flat per-team pricing, no Agent to install first. Join the early-access list, no card required.

No card required · your data stays yours

Keep exploring