AutoMerge is generally available

  • 42 min read

13 Best APM Tools in 2026: OpenTelemetry Support, Pricing and Trade-offs Compared

You're choosing an APM tool for one of a few reasons:

  • your current contract is up for renewal and the bill has outgrown the value
  • your team has decided to standardize on OpenTelemetry
  • your cloud provider is retiring the SDK you instrumented with
  • your services have multiplied and nobody can say where a slow request spends its time

Every major APM tool now says it supports OpenTelemetry, which makes vendor feature pages hard to tell apart.

The real differences are in how they want you to instrument your code, what still works once your data arrives through OpenTelemetry, and the unit they bill on. This article compares 13 APM tools on exactly those points. You'll leave with a short list that fits how you instrument, how you deploy and how your bill would grow. If you want the summary first, jump to the comparison table.

What an APM tool does

An application performance monitoring (APM) tool measures how your services behave in production. It tracks request rate, error rate and latency for every service and endpoint, and traces individual requests across services so you can see which call, query or code path made a request slow or made it fail. Most APM tools share a core set of features: a service map, RED metrics (rate, errors and duration), distributed tracing, error tracking, profiling, and visibility into database and external calls.

APM is the application-centric part of observability. Observability platforms add infrastructure metrics, logs and other signals around it, and most tools in this list sell both. For a deeper look at the capabilities themselves, see Application Performance Monitoring Tools: 4 Key Capabilities.

What changed in 2026

OpenTelemetry (OTel) is the open standard for instrumenting applications and exporting traces, metrics and logs. The Cloud Native Computing Foundation (CNCF) announced its graduation on May 21, 2026. Its trace, metric and log signals are stable at the protocol level (status), and the profiling signal is still in development.

The cloud providers are moving the same way:

  • AWS put the X-Ray SDKs and daemon into maintenance mode on February 25, 2026, with releases limited to security fixes. AWS recommends moving to OpenTelemetry through the AWS Distro for OpenTelemetry.
  • Microsoft says the classic .NET Application Insights SDK 2.x is deprecated and retires on March 31, 2027. New apps should use the Microsoft OpenTelemetry Distro.
  • Google Cloud recommends its OTLP-based Telemetry API over the older Cloud Trace API.

So "does it support OpenTelemetry?" no longer separates tools. Every vendor in this list accepts OTLP, the OpenTelemetry protocol. The questions that do separate them come next.

How to compare APM tools in 2026

Compare APM tools on three things, in this order:

  1. How you instrument.
  2. What works with that data.
  3. Whether the pricing unit fits your architecture.

Then check sampling and retention, because they decide what you can still investigate after an incident.

Instrumentation comes first because it's the most expensive thing to change later. Rebuilding dashboards is a small project; re-instrumenting dozens of services is a large one.

To be included here, a tool has to offer distributed tracing and service-level performance views. It also has to be maintained and available today, and document enough to answer the same questions as every other entry:

  • how it collects data
  • what's limited with OpenTelemetry data
  • which languages it auto-instruments
  • what its sampling and retention defaults are
  • how it prices
  • where it runs

Jaeger is the one exception: it's tracing only, and it's included as the open-source baseline the others build on. Two entries cover a product family rather than a single product: Splunk's two APM products share one, and the three cloud providers share another.

Instrumentation: vendor agent, OpenTelemetry or eBPF

The tools in this list fall into four instrumentation models:

  • Vendor agent. A language agent from the vendor, such as New Relic's agents or Datadog's tracers, instruments your code and sends data in the vendor's format. These are mature and often go deeper than OpenTelemetry today, especially for code-level detail. The cost is that switching vendors means re-instrumenting.
  • Vendor OpenTelemetry distribution. OpenTelemetry SDKs or a Collector packaged by the vendor, such as Elastic's EDOT, Splunk's OTel Collector distribution or Grafana Alloy. The instrumentation stays standard, with vendor defaults on top.
  • Upstream OpenTelemetry. The project's own SDKs, auto-instrumentation agents and Collector, sending OTLP to whichever backend you choose. Changing backends is mostly a configuration change.
  • eBPF. Kernel-level instrumentation that sees HTTP, database and network calls without touching your code. Coroot and Grafana Beyla work this way, and so does OpenTelemetry eBPF Instrumentation (OBI), which reached version 1.0 in late 2026. Coverage is broad, but traces can be less complete than SDK instrumentation.

Vendor agents can be the deeper option today. OpenTelemetry keeps your instrumentation portable. Which matters more depends on how likely you are to switch, and how much the vendor-only features are worth to you. For the open-source side of this ecosystem, see the best OpenTelemetry tools and what makes a good OpenTelemetry backend.

"Accepts OTLP" is not "full features with OTel data"

Every vendor below accepts OpenTelemetry data. Several of the largest also document, in their own docs, features that are limited or unavailable when your data arrives that way. Read the compatibility page before you sign, not after.

  • New Relic. OTLP is New Relic's preferred ingest method, and OpenTelemetry services get its curated APM UI. Its docs warn that cumulative metric temporality "results in high data ingest". For OpenTelemetry services, the APM UI's segment breakdown is extrapolated from sampled spans.
  • Splunk AppDynamics (self-hosted Virtual Appliance). Release 26.7.0, from August 2026, ended legacy direct OpenTelemetry ingestion. OTel data now has to go through an OTel gateway.
  • Elastic. Elastic's Managed OTLP Endpoint is the simplest way to send OTel data to Elastic. Its documented limitations for OpenTelemetry include no managed tail-based sampling and no Universal Profiling, which works only on the classic stack. Many prebuilt integrations and dashboards are designed for ECS-formatted data.
  • Datadog. Its OpenTelemetry compatibility page lists which features work with each setup.
    • With the OpenTelemetry SDK and the upstream Collector, Database Monitoring, Cloud Network Monitoring, Live Containers and Universal Service Monitoring aren't supported. They need either the Datadog Distribution of the OTel Collector (DDOT) or the Datadog SDK.
    • Sending OTLP straight to Datadog gives you sampled trace metrics, and no host list or Source Code Integration.
  • Sentry. Sentry accepts OTLP traces and logs, but doesn't support OTLP metrics.

If you're committed to one vendor and its agent covers your languages, the agent may still give you more today. If OpenTelemetry is your plan, these pages tell you which parts of the product you'd pay for and not get.

Pricing units and the costs vendors document

APM list prices are hard to compare because the units differ. Each unit penalizes a different shape of system. Check the unit before the number:

Pricing unitToolsGets expensive when you have…
Per hostSplunk Observability Cloud, Splunk AppDynamics, DatadogMany small hosts, or containers billed as hosts
Per GB ingestedNew Relic, Elastic, Grafana Cloud, SigNoz, AzureVerbose spans and attributes
Per userNew Relic (full platform users)A large engineering team that all needs access
Per event or spanHoneycomb (events), Sentry, Google Cloud Trace, Dash0 (spans)High request volume with little sampling
Per CPU coreCoroot StandardLarge clusters

Several vendors document costs that are easy to miss:

  • New Relic: data ingest and user seats are billed separately. Full platform users cost $349 per month on the Pro plan (annual) (pricing).
  • Grafana Cloud: Adaptive Traces does managed tail sampling, but "GBs processed" is billed before any trace is dropped.
  • Datadog:
    • If you run one Agent per container, each container is billed as a host.
    • Profiling more than four containers per host costs extra.
    • Each APM host includes 1 million indexed and 150 GB of ingested spans. Beyond that, you pay per million indexed spans by retention period.
    • The Datadog pricing guide breaks this down.

Sampling and retention

Sampling decides which traces are kept. Retention decides how long they stay searchable. If the trace you need was sampled away or expired before you looked, the APM tool can't help, however good its UI is.

ToolDefault sampling or captureTrace retention
New RelicAdaptive head sampling; optional Infinite Tracing tail sampling (keeps 1% by default)8 days (98 days with Data Plus)
Splunk Observability CloudFull-fidelity "NoSample" tracing8 days raw by default (15 or 30 available)
ElasticConfigurable; no managed tail sampling for OpenTelemetry dataConfigurable by retention pricing
DatadogDatadog SDKs: head sampling targets 10 traces/s per Agent, plus an error sampler at 10 traces/s (not applied to OTel SDKs)Indexed spans 15 days (custom filters); intelligent retention filter 30 days
Dash0Keeps what you send; optional tail sampling with SignalControl30 days
Grafana CloudAdaptive Traces (managed tail sampling) on all tiers30 days
HoneycombNo sampling by default; optional tail sampling with Refinery60 days
SigNoz CloudNot documented15 days
SentrySet by your SDK sample rate30 to 90 days by plan
AWS X-Ray1 request per second, plus 5% of additional requestsNot documented
Azure Application InsightsJava agent: 5 requests per second90 days included
Google Cloud TraceSet by your OpenTelemetry sampler30 days

For the self-hosted tools (Apache SkyWalking, Coroot and Jaeger), sampling and retention depend on how you configure them and the storage you run. Whatever you shortlist, check that its defaults reach as far back as your incidents usually do.

APM tools at a glance

ToolInstrumentationOpenTelemetry notesPricing unitEntry list price (Nov 2026)Free tierDeployment
New RelicVendor agents or OTLPOTLP is the preferred methodPer GB + per user$0.40/GB beyond free allowance100 GB/month, 1 full userSaaS
Splunk Observability Cloud / AppDynamicsSplunk OTel distribution; AppDynamics agentsAppD self-hosted needs an OTel gatewayPer host$60/host/month (App & Infra, annual)14-day trial (Observability Cloud)SaaS; AppDynamics also self-hosted
Elastic ObservabilityEDOT (OTel distribution)Managed endpoint limitsPer GB ingested + retainedIngest as low as $0.09/GBFree trialServerless, cloud, self-managed
DatadogVendor agent, DDOT or OTLPSome features need DDOT or the Datadog SDKPer host + indexed spansAPM from $31/host/month (annual)14-day trialSaaS
Dash0Upstream OTel; Dash0 Operator on KubernetesStores OTLP nativelyPer million spans, logs, data points$0.06 ingest + $0.54 storage per million spans14-day trialSaaS
Grafana CloudOTel SDKs + Alloy; Beyla eBPFOpenTelemetry-basedPer host hour + per GB$19/month + $0.025/host hour50 GB traces, 14-day retentionSaaS; Tempo self-hosted (AGPL-3.0)
HoneycombUpstream OTelOTLP-nativePer eventFrom $150/month per 50M events20M events/monthSaaS
SigNozUpstream OTelOTel-nativePer GB + per metric samples$49/month including usageSelf-hosted CommunityCloud; self-hosted (MIT core)
SentrySentry SDKs; OTLP traces and logsNo OTLP metricsPlan + spans and errorsTeam from $26/month (annual)Developer plan, 1 userSaaS; self-hosted (FSL)
Apache SkyWalkingNative agents; OTLP traces (v11.1+)Native OTLP traces since v11.1FreeFreeOpen sourceSelf-hosted (Apache-2.0)
CorooteBPF agent; OpenTelemetryAccepts OTel; eBPF traces may be partialPer CPU core (Standard)$1/core/monthCommunity editionSelf-hosted (Apache-2.0)
JaegerUpstream OTel (built on the Collector)Tracing onlyFreeFreeOpen sourceSelf-hosted (Apache-2.0)
AWS / Azure / Google CloudADOT; Microsoft OTel Distro; upstream OTelEach has its own OTLP endpointPer signal, GB or spanSee entryFree allowancesCloud SaaS

Quick picks:

  • Staying with a vendor agent: New Relic, Splunk or Datadog.
  • Standardizing on OpenTelemetry: Dash0, Grafana Cloud, Honeycomb or SigNoz.
  • Self-hosting: SigNoz, SkyWalking or Coroot.
  • Error-first workflow: Sentry.
  • One cloud: your cloud provider's APM.

The section on which tool fits your situation maps more situations.

Full-stack commercial platforms

These four cover APM alongside infrastructure, logs and user experience monitoring. Each leads with its own agent or OpenTelemetry distribution. Each also documents where OpenTelemetry data gets fewer features.

1. New Relic

New Relic offers its own language agents and native OTLP ingest over gRPC and HTTP, which it calls the preferred method. OpenTelemetry services show up in the same curated APM views as agent-instrumented ones. On Kubernetes, its agent operator injects agents for Java, .NET, Node.js, Python, Ruby and PHP.

What's good

  • A usable free tier: 100 GB of data per month and one full platform user.
  • OpenTelemetry treated as a first-class path into the APM UI.
  • AI features, including New Relic Autopilot.

The catch

Pricing combines data ingest with fees for full platform users, which adds up for large teams. Spans are retained for 8 days unless you pay for Data Plus. For OpenTelemetry services, segment breakdowns in the APM UI are extrapolated from sampled spans.

Pricing model

New Relic charges $0.40 per GB beyond the free 100 GB ($0.60 with Data Plus). Core users cost $49 per month. Full platform users are $10 for the first and $99 for each additional on Standard (maximum five). Pro is $349 per user per month billed annually.

The verdict

Worth exploring if: you want a generous free tier and an APM UI that handles OpenTelemetry data directly.

Give it a pass if: many engineers need full access, since seats can cost more than data.

See also: New Relic alternatives.

2. Splunk Observability Cloud and Splunk AppDynamics

Cisco now sells two APM products under the Splunk name. Splunk Observability Cloud is OpenTelemetry-based, through the Splunk Distribution of the OpenTelemetry Collector, with zero-code instrumentation for Java, Node.js, .NET, Go, Python, Ruby and PHP. Splunk AppDynamics is the long-established agent-based APM. In June 2026, Splunk said it had brought the best of AppDynamics SaaS into Observability Cloud, including business transactions. AppDynamics remains for self-managed and air-gapped customers.

What's good

  • Full-fidelity "NoSample" tracing in Observability Cloud.
  • Tag Spotlight for breaking down latency and errors by span attributes.
  • AlwaysOn Profiling for Java, .NET and Node.js.
  • A self-hosted option, through AppDynamics, for environments that can't use SaaS.

The catch

Two products with overlapping features means more decisions about which to run. The self-hosted AppDynamics Virtual Appliance no longer accepts direct OpenTelemetry ingest; data goes through an OTel gateway.

Pricing model

Splunk lists Observability Cloud's App & Infra tier, which includes APM, profiling and synthetic API checks, at $60 per host per month billed annually. End-to-End adds RUM and browser synthetics at $75. AppDynamics Premium is $33 and Enterprise is $50, both billed annually. The pricing page labels the AppDynamics unit both per host and per CPU core, so check the unit in your quote. Observability Cloud has a 14-day free trial.

The verdict

Worth exploring if: you already run Splunk or AppDynamics, or you need APM in a self-managed or air-gapped environment.

Give it a pass if: you want one clear product and a simple bill.

See also: AppDynamics alternatives.

3. Elastic Observability

Elastic Observability collects APM data through EDOT, the Elastic Distributions of OpenTelemetry. That's a Collector plus SDKs for Java, .NET, Node.js, PHP, Python, Android and iOS. Go and Ruby use upstream SDKs with community support. On Kubernetes, you use the upstream OpenTelemetry Operator with EDOT images. Elastic's Managed OTLP Endpoint is generally available on Serverless and Cloud Hosted 9.0+.

What's good

  • APM, logs and search in one place, built on Elasticsearch.
  • Run it as Serverless, on Elastic Cloud or self-managed.
  • AI Assistant and agent-building features.

The catch

The managed OTLP endpoint isn't available for self-managed deployments, and OpenTelemetry data has documented feature limits (see above). Many prebuilt views were designed for Elastic's own data format.

Pricing model

Elastic Serverless Observability Complete costs as low as $0.09 per GB ingested and $0.019 per GB-month retained. Support costs 5% to 15% of consumption. Self-managed licensing is by quote. A free trial is available.

The verdict

Worth exploring if: you already run Elasticsearch for logs or search, or you need a self-managed option.

Give it a pass if: you plan to send OpenTelemetry data and need managed tail sampling.

4. Datadog

Datadog APM is part of one of the broadest observability platforms on the market. You can instrument with Datadog's own tracing libraries and Agent, with DDOT, with the upstream OpenTelemetry Collector, or by sending OTLP straight to Datadog. Single Step Instrumentation auto-instruments services on Linux, Docker, Kubernetes and Windows hosts.

What's good

  • A very wide feature set around APM: Error Tracking, Continuous Profiler, Database Monitoring, Data Streams Monitoring, RUM and Synthetics.
  • Several ways to bring in OpenTelemetry data, including a supported Collector distribution.
  • AI features under the Bits brand, including Bits AI SRE.

The catch

The richest features assume Datadog's own SDK or DDOT, as described above. Billing combines hosts, ingested spans, indexed spans and per-product add-ons. Container-heavy setups need care, because an Agent per container is billed as a host.

Pricing model

Datadog lists APM at $31 per host per month billed annually ($36 on-demand). APM Pro is $35 and APM Enterprise, which includes the profiler, is $40. Ingested spans beyond the allowance cost $0.10 per GB. Indexed spans cost $1.70 per million at 15-day retention. Database Monitoring is $70 per database host. There's a 14-day free trial.

The verdict

Worth exploring if: you want one vendor for APM, infrastructure, logs and security, and you're comfortable instrumenting with Datadog's libraries.

Give it a pass if: you're standardizing on upstream OpenTelemetry, or you run many containers per host and need a predictable bill.

See also: Datadog alternatives.

OpenTelemetry-native platforms

These four are built around OTLP and upstream OpenTelemetry instrumentation rather than a vendor agent. They differ in pricing unit, scope and whether you can run them yourself.

5. Dash0

Dash0 is an OpenTelemetry-native observability platform. It stores OTLP traces, metrics, logs and profiles as they arrive, and computes RED metrics for every service and operation from your spans at ingestion. The Service Catalog and Service Map show service health and dependencies, Triage highlights the attribute values that set slow or failing spans apart from a baseline, and you query everything with PromQL or SQL.

On Kubernetes, the Dash0 Operator auto-instruments Java 8+, Node.js 16+ and .NET 6+, with Python and Ruby 3.3+ as opt-in. Continuous profiling became available on November 10, 2026, using an eBPF profiler the Operator deploys without code changes.

What's good

  • OpenTelemetry resource attributes and semantic conventions drive navigation.
  • RED metrics need no extra instrumentation, and are computed before sampling when you use SignalControl tail sampling.
  • eBPF profiling covers Go, Rust, C/C++, Java, Python, Node.js, .NET, Ruby, PHP and more.
  • Agent0, an AI SRE, investigates incidents on the same data.

The catch

Dash0 is SaaS only. The Operator's auto-instrumentation is Kubernetes-only, so elsewhere you set up OpenTelemetry SDKs and a Collector yourself. The profiler is opt-in and off by default, runs with privileged node access, and doesn't support GKE Autopilot yet. Errors show up through span status and RED metrics; there's no dedicated issue-grouping or database-monitoring product. Agent0 is billed separately.

Pricing model

Dash0 charges $0.06 per million spans to ingest plus $0.54 per million to store, with 30-day retention. Logs are priced the same. Metric data points cost $0.02 per million to ingest and $0.18 per million to store, with 13-month retention. There are no per-host or per-seat fees, and a 14-day free trial is available. Agent0 costs $0.60 per credit.

The verdict

Worth exploring if: you're standardizing on OpenTelemetry, run on Kubernetes, and want APM priced on telemetry volume rather than hosts or seats.

Give it a pass if: you need to self-host, or you want dedicated error-grouping and database-monitoring products.

6. Grafana Cloud Application Observability

Grafana Cloud Application Observability is built on OpenTelemetry SDKs, Grafana Alloy (Grafana's distribution of the OpenTelemetry Collector) and the Prometheus data model. Grafana ships its own distributions of the OTel Java and .NET SDKs; other languages use upstream SDKs. Beyla, Grafana's distribution of OpenTelemetry eBPF Instrumentation, covers services you can't or won't change. Organizations created from September 7, 2026 get a version built on Grafana's Knowledge Graph, which discovers services automatically and connects their metrics, logs, traces and profiles.

If you'd rather run it yourself, Grafana Tempo is the open-source trace store underneath. It's AGPL-3.0 licensed and needs only object storage. Tempo 3.0 shipped in 2026 with a new write architecture and generally available TraceQL metrics.

What's good

  • OpenTelemetry and Prometheus throughout, with eBPF for uninstrumented services.
  • Metrics, logs, traces and profiles in Grafana.
  • A real free tier, and a self-hosted path through Tempo.

The catch

Pricing combines host hours with per-GB telemetry charges, and changed for customers who started on or after February 13, 2026. Adaptive Traces bills "GBs processed" before sampling drops anything. Self-hosting the full stack means running several components.

Pricing model

Grafana Cloud Pro costs $19 per month, plus $0.025 per host hour for Application Observability (about $18 per host per month). Traces cost $0.05 per GB processed, $0.40 per GB written and $0.10 per GB retained, and Pro keeps traces for 30 days. The free tier includes 50 GB of traces and 2,232 host hours with 14-day retention. Enterprise starts at a $25,000 annual commitment.

The verdict

Worth exploring if: you already use Grafana and Prometheus, or you want a managed service with a self-hosted fallback.

Give it a pass if: you want a single pricing unit that's easy to forecast.

7. Honeycomb

Honeycomb ingests OTLP from any OpenTelemetry SDK or Collector and is built for exploring high-cardinality trace data. BubbleUp compares slow or failing requests with the rest and surfaces the attributes that differ. Its own Beelines SDKs and OTel distribution have reached end of life, so new instrumentation is upstream OpenTelemetry.

What's good

  • Fast, query-driven investigation of traces with many attributes.
  • No sampling by default, with Refinery for tail sampling when you need it.
  • 60-day retention for events.

The catch

Honeycomb is SaaS only, and its scope is narrower than full-stack platforms: there's no profiling or database-monitoring product. The Service Map is Enterprise-only, and SLOs require Pro or Enterprise.

Pricing model

Honeycomb's free plan covers 20 million events per month. Pro starts at $150 per month per 50 million events, up to 750 million. Enterprise is custom.

The verdict

Worth exploring if: your team debugs by asking new questions of trace data rather than reading fixed dashboards.

Give it a pass if: you want profiling, infrastructure and database monitoring in the same product.

8. SigNoz

SigNoz is an OpenTelemetry-native APM and observability platform backed by ClickHouse, available as a cloud service or self-hosted. It computes RED metrics and Apdex from entry spans, and shows database calls from spans that carry OpenTelemetry database attributes. It also has a service map, exception tracking and LLM observability.

What's good

  • OpenTelemetry end to end, with the same product in the cloud or on your own infrastructure.
  • An open-source core under the MIT licence.
  • Low entry price for the cloud service.

The catch

Self-hosting means running ClickHouse, ClickHouse Keeper (or ZooKeeper) and Postgres. SAML, anomaly detection and multiple ingestion keys are paid-only. Release 0.150.0 removed the v1 alert rules APIs, so self-hosters need to plan upgrades.

Pricing model

SigNoz Teams costs $49 per month including $49 of usage. Logs and traces cost $0.30 per GB, and metrics $0.10 per million samples. Traces are kept for 15 days. Enterprise starts at $4,000 per month. The self-hosted Community edition is free.

The verdict

Worth exploring if: you want OpenTelemetry-native APM that you can start in the cloud and move in-house later.

Give it a pass if: you don't want to operate ClickHouse and the cloud service's retention is too short.

Developer-first APM

Sentry starts from errors and the code that caused them, and adds tracing and profiling on top.

9. Sentry

Sentry uses its own SDKs to capture errors, traces, session replays, profiles and logs, and links an exception to the release and commit behind it. It also accepts OTLP traces and logs, but not OTLP metrics.

What's good

  • Error tracking with stack traces, release tracking and session replay, all linked to traces.
  • Low entry price and a free single-user plan.
  • A self-hosted option.

The catch

Sentry isn't a full-stack platform: there's no service map or Kubernetes auto-instrumentation, and OpenTelemetry metrics aren't supported. Self-hosted Sentry is heavy: Kafka, ClickHouse, Postgres, Redis and more, with 16 GB of RAM recommended. Its licence is FSL-1.1-Apache-2.0, not a traditional open-source licence. Seer, its AI feature, isn't available self-hosted.

Pricing model

Sentry's Developer plan is free for one user. Team costs $26 per month and Business $80 per month, both billed annually. Paid plans include 5 million spans; extra spans on Team cost $2.00 per million. Continuous profiling costs $0.0315 per hour.

The verdict

Worth exploring if: your team works error-first and wants exceptions, releases and traces in one developer tool.

Give it a pass if: you need service-level APM across a large microservices estate.

See also: Sentry alternatives.

Open-source and self-hosted APM

Open-source APM is free to license, but you run the storage underneath it. Choose by what you're willing to operate.

10. Apache SkyWalking

Apache SkyWalking is an Apache-2.0 APM with native auto-instrumentation agents for Java, C#, Node.js, Go, PHP and Python. It also has an eBPF agent (Rover) and tracing, eBPF and continuous profiling. Version 11.1.0 (November 2026) stores OTLP traces natively and derives service metrics and topology from them. As a result, OpenTelemetry-instrumented services no longer need a SkyWalking agent.

What's good

  • A complete, Apache-licensed APM with no commercial tier.
  • Profiling and browser monitoring built in.
  • Native OpenTelemetry trace support as of v11.1.

The catch

Version 11.0.0 removed the bundled web UI. The new Horizon UI ships separately and requires the v11 backend. The default storage is BanyanDB, with Elasticsearch or OpenSearch recommended at larger scale.

Pricing model

Free. There's no commercial tier.

The verdict

Worth exploring if: you need a full, self-hosted APM under a permissive licence.

Give it a pass if: you don't want to operate and upgrade the backend yourself.

11. Coroot

Coroot uses an eBPF node agent to collect metrics, logs, traces and profiles and build a service map with no code changes. It also accepts OpenTelemetry instrumentation. Its eBPF-based tracing covers HTTP, Postgres, MySQL, Redis, MongoDB and Memcached. The docs note that it "may not provide complete traces".

What's good

  • Service map and golden signals without instrumenting your code.
  • SLOs, log patterns and one-click profiling.
  • Apache-2.0 Community edition.

The catch

You run Prometheus (with remote write enabled) and ClickHouse. eBPF traces may be incomplete, so add OpenTelemetry where you need full traces. AI root-cause analysis isn't in the free Community edition.

Pricing model

Coroot Community is free. Standard costs $1 per monitored CPU core per month and adds SSO, RBAC and AI root-cause analysis.

The verdict

Worth exploring if: you want APM-style visibility across services you can't instrument quickly.

Give it a pass if: you need complete distributed traces from eBPF alone.

12. Jaeger (tracing baseline)

Jaeger is the CNCF's distributed tracing system, not a full APM. It's included here as the baseline for open-source tracing. Jaeger v2 is built on the OpenTelemetry Collector and accepts OTLP by default. Its storage options include Badger, Cassandra, ClickHouse, Elasticsearch and OpenSearch. Its Monitor tab can show RED metrics through the SpanMetrics connector. Jaeger v1 reached end of life on December 31, 2025.

What's good

  • A standard, Apache-2.0 trace backend that speaks OTLP natively.
  • Several storage backends to choose from.

The catch

Jaeger is tracing only: no logs, profiling, RUM or alerting. For APM you'd pair it with metrics and alerting tools.

Pricing model

Free.

The verdict

Worth exploring if: you only need to store and search traces, or you're building your own stack.

Give it a pass if: you want an APM product rather than a component.

See also: open-source distributed tracing tools.

What self-hosting actually requires

"Free" self-hosted APM means running a database, usually ClickHouse. This is what each option asks you to operate:

ToolWhat you runLicence
SigNozClickHouse, ClickHouse Keeper or ZooKeeper, PostgresMIT core; enterprise features separately licensed
UptraceClickHouse, PostgreSQL, RedisAGPL-3.0
CorootPrometheus, ClickHouseApache-2.0
Sentry (self-hosted)Kafka, ClickHouse, Postgres, Redis, Snuba, RelayFSL-1.1-Apache-2.0
Grafana TempoObject storageAGPL-3.0
Apache SkyWalkingBanyanDB (default), or Elasticsearch/OpenSearchApache-2.0
JaegerBadger, Cassandra, ClickHouse, Elasticsearch or OpenSearchApache-2.0

If the "relatively hassle-free" part matters more than the "self-hosted" part, a managed service built on the same open-source project, such as SigNoz Cloud or Grafana Cloud, is often the better trade.

Cloud-provider APM

If you run mostly on one cloud, its native APM is the easiest place to start. All three now point you to OpenTelemetry.

13. AWS, Azure and Google Cloud

AWS CloudWatch Application Signals and X-Ray

CloudWatch Application Signals uses the AWS Distro for OpenTelemetry and the CloudWatch agent to auto-instrument Java, Python, Node.js and .NET on EKS, ECS and EC2. It shows per-service volume, latency, faults and errors, and adds SLOs, an application map and change events from CloudTrail. CloudWatch accepts OTLP traces, metrics and logs over HTTP; gRPC isn't supported. Pricing, from AWS's own examples:

  • Application Signals: $1.50 per million signals for the first 100 million.
  • Span ingestion: $0.35 per GB.
  • X-Ray: $5 per million traces recorded, after the first 100,000 each month.
  • A free tier covers new usage.

(pricing)

Azure Monitor Application Insights

Application Insights uses the Microsoft OpenTelemetry Distro for .NET, Node.js and Python, and a no-code Java agent. It provides an application map, profiler and live metrics. OTLP ingestion through the OpenTelemetry Collector became generally available in June 2026; the Azure Monitor Agent and AKS ingestion paths are still in preview. If you send OTLP from your own Collector, Application Insights experiences need delta temporality and exponential histograms. Ingestion costs $2.30 per GB in East US after the first 5 GB per month. Application Insights data is retained 90 days at no extra charge.

Google Cloud Trace

Cloud Trace takes upstream OpenTelemetry data through Google's OTLP-based Telemetry API, with no proprietary agent. Traces are kept for 30 days. Observability Analytics, which became generally available in May 2026, lets you query traces with SQL. Spans cost $0.20 per million, and the first 2.5 million each month are free.

The verdict

Worth exploring if: you're all-in on one cloud and want the lowest-friction start.

Give it a pass if: you run across clouds or on-premises, or want one APM for everything.

Also considered

These tools are viable but didn't get a full entry:

  • IBM Instana: agent-based APM with 1-second metrics and OTLP support, sold as SaaS or self-hosted. Read Instana alternatives for how it compares.
  • Uptrace: OpenTelemetry-native and ClickHouse-based, AGPL-3.0. Smaller than SigNoz, with similar scope.
  • OpenObserve: an AGPL-3.0 observability platform with APM features. Its service graph is an Enterprise feature.
  • Middleware: OpenTelemetry-based SaaS with a large free allowance. Its documentation was inconsistent in places.
  • Site24x7 APM Insight: agent-based APM inside a broader IT monitoring suite.
  • ManageEngine Applications Manager: self-hosted application monitoring. OpenTelemetry support covers traces only so far.

Which tool fits your situation

Start from your situation. Each one points to a short list from the entries above.

  • You're deep in one vendor's agents and happy with the depth. Stay, but read the vendor's OpenTelemetry compatibility page before you add OTel services. Look at New Relic, Splunk or Datadog.
  • You're standardizing on OpenTelemetry and don't want to re-instrument again. Choose a backend that treats OTLP as its native format: Dash0, Grafana Cloud, Honeycomb or SigNoz. Elastic and Splunk's OTel distributions are also options if you already use those platforms.
  • You run many small containers on Kubernetes. Avoid units that turn containers into hosts. Look at tools priced on telemetry volume or cores, such as Dash0, Honeycomb, SigNoz or Coroot. Dash0 and Grafana also auto-instrument or use eBPF inside the cluster.
  • A large team needs access. Check per-user fees before data fees. New Relic charges for full platform users. Dash0 documents no per-seat fee.
  • You must self-host or run air-gapped. Look at SigNoz, Apache SkyWalking, Coroot, Grafana Tempo or Splunk AppDynamics, and budget for running the storage.
  • You're a small team that wants a free start. New Relic (100 GB per month), Grafana Cloud's free tier, Honeycomb (20 million events), Sentry's Developer plan and self-hosted SigNoz all let you start without a contract.
  • You're all-in on one cloud. Start with Application Signals, Application Insights or Cloud Trace, and instrument with OpenTelemetry so you can move later.
  • Your team works error-first. Sentry, possibly alongside a service-level APM.

FAQ

What is an APM tool?

An APM tool monitors how your applications perform in production. It tracks request rate, errors and latency per service. It traces requests across services and points you to the code, query or dependency behind a slowdown or failure.

What is the best APM tool?

There isn't one best tool for every team. The right choice depends on how you instrument, whether you need to self-host, and which pricing unit fits your architecture. If you're standardizing on OpenTelemetry, start with OTel-native backends. If you're committed to one vendor's agents, compare that vendor's depth against its bill.

Are open-source APM tools enough for production?

Yes, many teams run SigNoz, SkyWalking, Coroot or Grafana Tempo in production. The cost moves from licences to operations: you run and scale storage such as ClickHouse, handle upgrades, and own availability. Managed versions of the same projects remove that work.

What's the difference between APM and observability?

APM focuses on applications: services, requests, traces and code-level performance. Observability is broader. It covers infrastructure, logs and other signals, so you can investigate problems you didn't anticipate. Most APM vendors now sell both.

Does OpenTelemetry replace an APM agent?

For instrumentation, increasingly yes. OpenTelemetry SDKs and auto-instrumentation agents cover the major languages, and every vendor here accepts OTLP. Some vendors still reserve features for their own agents, so check their compatibility documentation. OpenTelemetry doesn't replace the backend: you still need a tool to store, query and alert on the data.

How long does switching APM tools take?

It depends mostly on how you instrumented. If your services already use OpenTelemetry, switching backends can be as small as changing the exporter endpoint in your Collector. Running both side by side during evaluation is straightforward. If you're on a vendor's proprietary agents, plan for re-instrumenting service by service. Expect to rebuild dashboards and alerts in either case.

Final thoughts

An APM tool is hard to swap out later, so build your shortlist on the decisions that are hardest to reverse, in the order set out in how to compare APM tools. Leave dashboards and AI features for the trial, where you can judge them on your own services.

For related comparisons, see the best observability tools, all-in-one observability platforms and distributed tracing tools.

If you're standardizing on OpenTelemetry, Dash0 stores your OTLP data natively. It derives RED metrics from your spans and prices on telemetry volume rather than hosts or seats. Start a 14-day free trial, add it as an exporter in your Collector, and compare it with the other tools on your shortlist using your own services. No credit card required.

Dash0 HTTP requests view showing latency outliers, service-level HTTP status codes and trace details for a selected 404 response.