Datadog LLM Observability Alternatives (2026): 8 Tools Compared

The best Datadog LLM Observability alternatives in 2026 are Langfuse if you want open-source tracing at a fraction of the cost, Arize Phoenix or SigNoz if you want OpenTelemetry-native tools you can self-host, Braintrust if your team works in evals and release gates, LangSmith if you build on LangChain or LangGraph, and Grafana Cloud, New Relic, or Dynatrace if you want to stay on a full-stack observability platform. If what you need is not another place to watch agent traces but a short list of what to fix in each workflow, Papaya does that with no configuration.

A naming note first. Datadog renamed LLM Observability to Agent Observability in June 2026, around its DASH conference. The docs still live under the old URL and most people still search for the old name, so we use both here. We checked every price and feature below against each vendor’s own pages on October 21, 2026.

Why teams look for a Datadog LLM Observability alternative

Datadog’s agent product is capable. It traces LLM calls from Python, Node.js, Java, and Go, accepts OpenTelemetry GenAI spans, runs managed and custom LLM-as-a-judge evaluations, versions datasets for experiments, clusters production traffic into topics, and surfaces cost and reliability Insights. Its real advantage is correlation: agent traces sit next to your APM, logs, and infrastructure. Teams that leave usually do so for one of four reasons.

Cost at agent scale. Datadog bills per LLM span. On annual billing the first 100,000 LLM spans a month cost $160, then $3.50 per 10,000 at the default 15-day retention. Datadog counts only the model calls, not tool or retrieval spans, which helps. But agents make many model calls per run, so span counts grow much faster than users or conversations.

Short retention. Traces are kept for 15 days by default. You can pay for 30, 60, or 90 days, and 90 is the maximum. If you want to compare this quarter’s agent behavior with last quarter’s, that is a hard limit.

No self-hosting. Datadog is proprietary SaaS. Your prompts and responses live in Datadog’s hosted regions, and the product is not available on Datadog’s US-FED site. Teams with strict data-residency rules or an open-source policy look elsewhere.

Built around APM. The SDK is built on Datadog’s APM tracer, and the product shines when you already pay for Datadog. If your observability lives elsewhere, or you only need agent visibility, buying into the Datadog ecosystem for one product is a lot of platform for the job. It can be bought standalone, but most of its value is in the correlation.

Datadog LLM Observability alternatives at a glance

What it costs at agent scale: a worked example

Take a support agent that handles 20,000 conversations a month. Each conversation makes 7 model calls and 4 tool calls, the shape of a typical booking-change flow. That is 140,000 LLM spans and 220,000 spans in total.

Tool and plan	How it meters this workload	Approximate monthly cost
Datadog, annual, 15-day retention	140,000 LLM spans: $160 for the first 100,000, plus 4 × $3.50	About $174
Langfuse Core, 90-day retention	About 240,000 units (traces plus observations): $29 for 100,000, plus 1.4 × $8	About $40
Opik Pro, 60-day retention	220,000 spans: $19 for 100,000, plus 1.2 × $5	About $25

These are list prices as of October 21, 2026, before discounts, and they cover tracing only. The gap widens with longer retention, because Datadog’s 30-, 60-, and 90-day options cost more per span. It narrows if you already have a Datadog commitment. Treat the numbers as a way to compare metering, not as a quote.

The alternatives, by what they are best for

Papaya: best if you want fixes, not another dashboard

Datadog shows you what your agent did. Papaya tells you what to change. There is nothing to configure: you send traces, through Papaya’s one-line SDK wrapper or as exports from the tools you already use, and Papaya works out the structure itself. It classifies every run into the business use case it served, compares succeeding and failing runs with 200+ research-backed analyses, and ranks each finding by its expected impact on quality, latency, and cost, with pull requests for approved fixes. Its trace viewer pairs a clean detailed view of every run with a summary view of the whole workflow, so engineers and leadership can both read it. You can keep Datadog for infrastructure and APM and run Papaya alongside it. More on how it differs.

Langfuse: best open-source replacement

Langfuse is the closest like-for-like swap for teams that mostly use Datadog to trace LLM calls. It covers tracing, LLM-as-a-judge and code evaluators, datasets, experiments, prompt management, and dashboards, with an MIT-licensed core you can self-host for free. Cloud plans start at $29 a month with unlimited users and 90-day retention. ClickHouse acquired Langfuse in January 2026 and says the core stays MIT. It does not scan production for recurring failures on its own. See Langfuse alternatives for how it compares with the rest of the field.

Arize Phoenix and AX: best OpenTelemetry-first option

Phoenix is free to self-host, built on OpenTelemetry and OpenInference, and covers tracing, evals, datasets, experiments, and prompt iteration under the Elastic License 2.0, which is source-available rather than OSI open source. Arize AX adds hosted features from $50 a month and Signal, a scheduled scan that groups recurring failures. Note that Arize is now part of an observability company too: Dynatrace completed its acquisition on October 1, 2026, and says Phoenix and AX will continue. See Arize Phoenix alternatives.

Braintrust: best if quality gates matter more than monitoring

Braintrust is built around evaluation: production logs and experiments share one data model, so a bad trace becomes a test case quickly, and you can gate releases on scores. It has online scoring, alerts, an OpenTelemetry endpoint, and Topics for clustering traces. It is closed source, Pro is $249 a month, and self-hosting is Enterprise only. See Braintrust alternatives.

LangSmith: best for LangChain and LangGraph teams

LangSmith offers tracing, online evaluations, annotation queues, and datasets, plus Engine, which clusters failures into prioritized issues and can propose pull requests. Pricing is per seat ($39 a month on Plus) with 14-day retention on base traces, and self-hosting is an Enterprise add-on. See LangSmith alternatives.

Opik: best low-cost open-source platform

Opik, from Comet, is Apache-2.0 licensed and self-hostable, with trace trees, datasets, experiments, LLM-as-a-judge metrics, guardrails, and an agent optimizer. At $19 a month for 100,000 spans, it is the cheapest hosted option here.

SigNoz and OpenLIT: best if you want OpenTelemetry everywhere

If what you like about Datadog is one place for all telemetry, SigNoz is an open-source, OpenTelemetry-native platform that accepts any gen_ai.* instrumentation, including OpenLLMetry, OpenLIT, and OpenInference, and adds an AI observability view. OpenLIT is an Apache-2.0 SDK and platform for LLM tracing, evaluations, guardrails, and GPU monitoring that emits standard OTLP, so you can send the same data to any backend.

Grafana Cloud, New Relic, Dynatrace, and Honeycomb: if you want to stay full-stack

If you are leaving Datadog as a whole rather than just its agent product, the other full-stack platforms all cover LLM traces now. Grafana Cloud’s AI observability is built on OpenLIT, with a free tier of 30,000 generations a month. New Relic made OpenTelemetry GenAI support generally available in August 2026, with no New Relic agent required. Dynatrace’s AI Observability ingests OpenTelemetry and OpenLLMetry and now owns Arize. Honeycomb groups agent spans into an agent timeline using OpenTelemetry conventions. All four make agent traces visible next to the rest of your system, which is Datadog’s main strength, but none of them is built to tell you what to fix in an agent.

Two tools that often appear on older lists need a caveat. Helicone was acquired by Mintlify in March 2026 and is in maintenance mode. Galileo was acquired by Cisco and is now Splunk Agent Observability.

When Datadog is still the right call

Stay on Datadog if you already run Datadog APM and logs, your on-call engineers live in it, and the main question you ask about your agent is how it behaves next to the rest of your system. Following one slow request from the browser through your API into an agent run and back is something Datadog does better than any tool built only for agents. Its evaluations, topic clustering, and Insights are real features, and buying it standalone is possible. The case for switching is strongest when span costs, retention, self-hosting, or a non-Datadog stack matter more than that correlation.

Papaya: what to fix, not just what happened

Every tool above, Datadog included, starts from visibility. Papaya starts from the decision you actually need to make: what to change in your agent, and in which workflow.

Zero configuration. You send your traces and Papaya does the rest. You do not write detectors or scorers, map your schema, tag workflows, or tell it where things live in your traces. It works out the structure of your data and which business use cases your agent serves by itself. Your part is mostly giving feedback on what it finds.

A trace viewer anyone on the team can read. The detailed view shows the full trace, formatted cleanly from whatever format you sent. The summary view shows the high-level workflow, the shape of the agent step by step. Engineers zoom straight into the step where things go wrong, and leadership can see what an agent actually does without reading JSON.

Ranked fixes. Within each business use case Papaya compares succeeding and failing runs and runs 200+ research-backed analyses to find what separates them: tool calls that keep failing, steps that lead to hallucinations, sub-agents that bloat the context, and use cases that are not working at all. Each finding is quantified, so you fix the most valuable ones first, and with code access Papaya opens the pull request. We explain the difference between watching and fixing in agent optimization vs. observability, and compare the wider field in the best AI agent observability tools.

Try it on your own traces

Papaya’s free plan covers one agent and 1,000 traces a month. See pricing or talk to us.

Frequently asked questions

What is the best alternative to Datadog LLM Observability?

Langfuse is the best open-source replacement for tracing LLM calls at much lower cost. Arize Phoenix and SigNoz suit OpenTelemetry-first teams that want to self-host. Braintrust suits evaluation-driven teams and LangSmith suits LangChain users. Grafana Cloud, New Relic, and Dynatrace suit teams staying on a full-stack platform. Papaya suits teams that want ranked fixes per business use case with no configuration.

Did Datadog rename LLM Observability?

Yes. In June 2026 Datadog renamed LLM Observability to Agent Observability. The documentation still lives under the llm_observability path, and the pricing and features stayed the same at the time of the rename.

How much does Datadog LLM Observability cost?

As of October 2026, Datadog Agent Observability includes a free tier of 40,000 LLM spans a month. Paid use costs $160 a month on annual billing for the first 100,000 LLM spans, then $3.50 per 10,000 at 15-day retention, with higher rates for 30-, 60-, and 90-day retention. Only LLM spans are billed, not tool or retrieval spans.

Is there an open-source alternative to Datadog LLM Observability?

Yes. Langfuse (MIT core), Opik (Apache-2.0), OpenLIT (Apache-2.0), and SigNoz’s Community Edition are open source and self-hostable. Arize Phoenix is free to self-host under the Elastic License 2.0, which is source-available rather than OSI open source.

Can I use Papaya alongside Datadog?

Yes. Papaya does not replace your infrastructure or APM monitoring. You can keep Datadog for that and send agent traces to Papaya through its SDK or as exports, which Papaya maps automatically with no configuration, to get ranked fixes per business use case.

Sources used in this article

• Datadog pricing list, Agent Observability documentation, and OpenTelemetry instrumentation.
• Langfuse pricing and Langfuse joins ClickHouse.
• Arize pricing and Dynatrace completes acquisition of Arize.
• Braintrust pricing and LangSmith pricing.
• Comet Opik pricing.
• SigNoz LLM observability, SigNoz pricing, and OpenLIT on GitHub.
• Grafana Cloud pricing, New Relic OpenTelemetry AI monitoring, Dynatrace AI Observability, and Honeycomb agent observability.
• Mintlify acquires Helicone and Splunk on the Galileo acquisition.