Prices verified 30 October 2026

Langfuse alternatives, compared honestly.

Langfuse is the open-source default for LLM tracing, with a cheap hosted tier and a large community. Teams look elsewhere when the unit count makes the bill hard to predict, when 90 days of data access is not enough, when running it themselves stops being worth the engineering time, or when they need production alerting that tracing does not provide. The ClickHouse acquisition in January 2026 added a fifth reason for some. These nine tools each answer at least one of those. Prices were read from each vendor's pricing page on 30 October 2026 unless a different date is shown.

Agent Ping is on this list and is our product; it is marked as such. Every price links to the vendor's own page. Corrections welcome.

Why teams look for a Langfuse alternative.

Every observation is a unit.

A trace, an observation and a score each count as one unit, so a tool-using agent run is ten or more and a deep-research run is forty-plus. The bill moves with how richly you instrument, which is the opposite of what a budget wants.

Data access is 90 days on Core.

Three years of access needs Pro at $199 a month. Debugging a complaint from a quarter ago on Core means the trace is gone.

Self-hosting is free, running it is not.

Langfuse ships Docker Compose and Kubernetes templates and runs on ClickHouse, Postgres, Redis and object storage. That is four services to operate and upgrade for an internal tool.

Tracing is not operations.

Langfuse will show you what a run did. It will not page you when a scheduled agent stops running, score every production run against a rubric without you building the pipeline, or give you a flat bill.

It is now part of ClickHouse.

The product stays MIT-licensed and Langfuse Cloud continues, and the team says it will keep building. The roadmap is nonetheless set inside a database company, which some buyers want to price in.

The 9 alternatives at a glance.

Tool Billing unit Hosting Best for
LangSmith Traces: one end-to-end run tree, however many calls it contains Hosted; self-hosting and hybrid only on Enterprise Teams inside the LangChain and LangGraph ecosystem whose main loop is offline evaluation and prompt iteration.
Braintrust Gigabytes of processed data plus scores Hosted; on-prem or hosted deployment on Enterprise Teams who write their own scorers and want evaluation results beside the trace, with unlimited seats.
Agent Ping (ours) Runs, within a flat monthly allowance Hosted; SDKs for Python, TypeScript, Go and Laravel are open source, and OpenTelemetry is accepted Teams running agents for paying customers who need to know what each run cost, whether the scheduled ones ran, and whether the output is still good, without building the evaluation pipeline.
Arize Phoenix and AX Trace spans and gigabytes Phoenix open source and local-first; AX hosted or self-hosted on Enterprise Teams already on OpenTelemetry who want a credible open-source tracing and evaluation layer they can run locally.
Laminar Gigabytes of trace data after compression Source on GitHub; on-premise only on Enterprise Small teams that want cheap, framework-agnostic tracing with search and a low monthly floor.
LangWatch Events: every LLM call, tool call, retrieval, evaluation or simulation step Apache 2.0 open source; self-host with Docker Compose on your own ClickHouse European teams that want an open-source platform with simulation-based testing and a per-seat price they can predict.
Pydantic Logfire Spans, logs and metrics (records) Hosted; the SDK is open source Python teams on Pydantic AI or FastAPI who want one observability tool for the whole application, not just the model calls.
Lunary Events Hosted; self-hosting listed as an Enterprise option Small teams that want prompt management and human review alongside traces at a low per-user price.
Datadog LLM Observability LLM spans; tool, retrieval and agent spans are free Hosted Teams already running Datadog who want model calls in the same pane as their APM and logs.

Helicone is left out because it has been in maintenance mode since Mintlify acquired it in March 2026, and AgentOps because its pricing page could not be verified on the date above.

Each one in detail.

Ordered by how often each comes up as the replacement, not by our preference. Pricing is quoted as published on the date shown; the caveat is the thing a buyer most often learns after signing up.

1. LangSmith

LangChain's tracing, evaluation and prompt platform, wired into LangChain and LangGraph by default.

Pricing
Developer free for one seat with 5,000 base traces a month; Plus $39 a seat a month with 10,000 base traces; Enterprise custom. Usage is metered in LSUs at $1 each: base traces work out at $5 per 1,000, extended retention at $7.50.
Verified
30 October 2026, www.langchain.com
Hosting
Hosted; self-hosting and hybrid only on Enterprise
Billing unit
Traces: one end-to-end run tree, however many calls it contains
Best for
Teams inside the LangChain and LangGraph ecosystem whose main loop is offline evaluation and prompt iteration.
Caveat
Seats and per-trace usage compound. A million traces a month at base retention is about $5,000 before seats, and the included trace allowance does not scale with the number of seats you pay for.

Agent Ping vs LangSmith →

2. Braintrust

Eval-first platform with CI quality gates and the most generous free tier in the category.

Pricing
Starter free with 1 GB of processed data and 10,000 scores a month; Pro $249 a month with 5 GB and 50,000 scores; Enterprise custom. Overage $4 per GB on Starter, $3 on Pro; unlimited users on every plan.
Verified
30 October 2026, www.braintrust.dev
Hosting
Hosted; on-prem or hosted deployment on Enterprise
Billing unit
Gigabytes of processed data plus scores
Best for
Teams who write their own scorers and want evaluation results beside the trace, with unlimited seats.
Caveat
The jump from free to $249 is steep for a small team, and production alerting and operations are secondary to the eval workflow. Starter keeps 14 days of data.

Agent Ping vs Braintrust →

3. Agent Ping our product

Production observability built around the agent run: cost per customer, heartbeat monitoring and quality scoring on a flat tier. This is our product.

Pricing
Free plan with no card; Starter $99, Team $199 and Business $399 a month, each with named limits, no metered billing and unlimited seats.
Verified
30 October 2026, agentping.io
Hosting
Hosted; SDKs for Python, TypeScript, Go and Laravel are open source, and OpenTelemetry is accepted
Billing unit
Runs, within a flat monthly allowance
Best for
Teams running agents for paying customers who need to know what each run cost, whether the scheduled ones ran, and whether the output is still good, without building the evaluation pipeline.
Caveat
Not an offline eval workbench: there are no datasets or prompt playgrounds, and the quality scoring is rubric-based on live runs rather than dataset-driven. Teams whose main loop is development-time evaluation will want one of the tools above beside it.

Agent Ping pricing → Agent Ping vs Langfuse →

4. Arize Phoenix and AX

Phoenix is the open-source, OpenTelemetry-based tracing and eval tool; Arize AX is the hosted platform above it.

Pricing
Phoenix is free to run yourself. Arize AX Free has 25,000 trace spans and 1 GB a month with 15 days retention; AX Pro is $50 a month with 50,000 spans, 10 GB and 30 days; AX Enterprise is custom with self-hosting.
Verified
30 October 2026, arize.com
Hosting
Phoenix open source and local-first; AX hosted or self-hosted on Enterprise
Billing unit
Trace spans and gigabytes
Best for
Teams already on OpenTelemetry who want a credible open-source tracing and evaluation layer they can run locally.
Caveat
The AX Pro span allowance is small for production agents, and the step above it is an Enterprise conversation.

5. Laminar

Trace ingestion, search and evals for AI agents, priced by compressed data volume.

Pricing
Free with 1 GB and 7 days of data access; Starter $30 a month with 3 GB and $2 per extra GB; Pro $150 a month with 10 GB and $1.50 per extra GB; Enterprise custom.
Verified
30 October 2026, laminar.sh
Hosting
Source on GitHub; on-premise only on Enterprise
Billing unit
Gigabytes of trace data after compression
Best for
Small teams that want cheap, framework-agnostic tracing with search and a low monthly floor.
Caveat
Data access is 7 days on the free plan and 30 days on Starter. The analysis features (Signals) are metered separately by the tokens Laminar's own models consume.

6. LangWatch

LLM engineering platform with agent simulation testing, evals, observability and a gateway.

Pricing
Developer free with 50k events a month and 14 days of data; Growth 29 euros per core seat a month with 200k events included, then 5 euros per 100k; Enterprise custom.
Verified
30 October 2026, langwatch.ai
Hosting
Apache 2.0 open source; self-host with Docker Compose on your own ClickHouse
Billing unit
Events: every LLM call, tool call, retrieval, evaluation or simulation step
Best for
European teams that want an open-source platform with simulation-based testing and a per-seat price they can predict.
Caveat
Per-seat pricing returns at the Growth tier, and every tool call is a billable event, so agentic workloads consume the allowance quickly.

7. Pydantic Logfire

General OpenTelemetry observability from the Pydantic team, with first-class Pydantic AI support.

Pricing
Personal free with 10 million spans, logs or metrics a month for one seat; Team $49 a month for five seats; Growth $249 a month with unlimited seats. Every plan includes 10 million records, then $2 per extra million.
Verified
30 October 2026, pydantic.dev
Hosting
Hosted; the SDK is open source
Billing unit
Spans, logs and metrics (records)
Best for
Python teams on Pydantic AI or FastAPI who want one observability tool for the whole application, not just the model calls.
Caveat
It is general observability with AI support rather than an agent operations tool: no run or goal semantics, no built-in judge, dashboards and alerts are yours to assemble. Retention is 30 days below Growth.

8. Lunary

LLM observability and prompt management with a free OpenAI tokenizer that outranks its product pages.

Pricing
Team $20 a user a month with 50,000 events, 300 playground queries and a year of history; overage $10 per 50k events. Enterprise custom, with self-hosting, SSO and PII masking.
Verified
30 October 2026, lunary.ai
Hosting
Hosted; self-hosting listed as an Enterprise option
Billing unit
Events
Best for
Small teams that want prompt management and human review alongside traces at a low per-user price.
Caveat
Cloud workspaces require payment before activation, and self-hosting is gated to Enterprise.

9. Datadog LLM Observability

LLM spans inside the Datadog console, correlated with the rest of your infrastructure.

Pricing
Free up to 40,000 LLM spans a month with 15 days retention. Pro from $160 a month on annual billing for 100,000 LLM spans, then $3.50 to $5 per additional 10,000 spans depending on billing mode.
Verified
23 October 2026, www.datadoghq.com
Hosting
Hosted
Billing unit
LLM spans; tool, retrieval and agent spans are free
Best for
Teams already running Datadog who want model calls in the same pane as their APM and logs.
Caveat
The most expensive option at scale, and it does not judge whether an answer was any good. Extended retention is a paid add-on per 10,000 spans.

Agent Ping vs Datadog LLM Observability →

How to choose.

What is the best open-source alternative to Langfuse?
Arize Phoenix for OpenTelemetry-based tracing and evaluation that runs locally, and LangWatch (Apache 2.0) for a fuller platform with simulation testing. Both are self-hostable. Laminar publishes its source on GitHub but gates on-premise deployment to Enterprise.
Is Langfuse still open source after the ClickHouse acquisition?
Yes. ClickHouse stated on 16 January 2026 that Langfuse remains open source under its MIT licence for core features and that Langfuse Cloud continues as a standalone service. The Langfuse team joined ClickHouse and says it will keep building the product.
Which Langfuse alternative is cheapest?
For evaluation-heavy work, Braintrust's free tier (1 GB and 10,000 scores a month). For general observability at volume, Pydantic Logfire at $2 per million records. For production agent monitoring, a flat tier such as Agent Ping ($99 to $399 a month with named limits) removes the unit count entirely. Langfuse self-hosted remains free if you have the people to run it.
Do I have to leave Langfuse to get production alerting?
No. Several teams keep Langfuse for traces and add a production tool beside it. Agent Ping accepts OpenTelemetry and its SDKs sit alongside an existing Langfuse integration; the migration guide covers both the replace and the run-both options.

More comparisons, and the pricing posts behind them.

Langfuse vs LangSmith → LangSmith alternatives → LangSmith pricing explained → Langfuse pricing explained → Braintrust pricing explained →