Both offer a free tool you can run yourself and a hosted platform, both are built on OpenTelemetry, and both were acquired in 2026. The difference that matters is what the free self-hosted version does: Langfuse's includes online evaluation of production traffic; Arize Phoenix points you to the paid AX platform for that.[3][6]
Both changed owners this year: ClickHouse acquired Langfuse in January 2026 and Dynatrace completed its acquisition of Arize on 1 October 2026. Langfuse is MIT-licensed and its free self-hosted version scores production traffic. Arize's free self-hosted tool, Phoenix, is source-available under the Elastic License, and continuous production evals sit in its paid platform, AX.
| Decision factor | Arize | Langfuse |
|---|---|---|
| Ownership | Acquired by Dynatrace, completed 1 October 2026; Phoenix and AX continue | Acquired by ClickHouse, January 2026; no planned licence changes |
| Products | Phoenix (self-hosted) and Arize AX (hosted or Enterprise self-hosted) | One product, hosted (Langfuse Cloud) or self-hosted |
| Licence | Phoenix: Elastic License 2.0 (source-available); AX: commercial | MIT, except add-ons in the ee folders |
| Online evals in the free self-hosted version | No; Phoenix docs point to AX for continuous production evals | Yes, evaluator rules with sampling |
| OpenTelemetry | Built on it; Arize maintains the OpenInference conventions | OTLP over HTTP; gRPC not supported |
| Runtime guardrails | Partial: AX guards run in your code via Guardrails AI | None; evaluation happens after the fact |
| Free hosted tier | AX Free: 25k spans a month, 15-day retention | Hobby: 50k units a month, 30 days of data |
| First paid tier | AX Pro: $50 a month, 50k spans | Core: $29 a month, 100k units |
| Capability | Arize | Langfuse | Prefactor |
|---|---|---|---|
| Observe | |||
| LLM call tracing | ✓ | ✓ | ✓ |
| Multi-step agent tracing (tool calls, nested spans) | ✓ | ✓ | ✓ |
| Cost tracking | ✓ | ✓ | ✓ |
| OpenTelemetry support | ✓ | ✓ | ✓ |
| Evaluate | |||
| Offline evals (datasets, experiments) | ✓ | ✓ | ✓ |
| Online evals on production traffic | ✓ | ✓ | ✓ |
| LLM-as-a-judge scoring | ✓ | ✓ | ✓ |
| Code-based evaluators | ✓ | ✓ | ✓ |
| Human review and annotation | ✓ | ✓ | ✓ |
| Prompt versioning and playground | ✓ | ✓ | — |
| Act on findings | |||
| Runtime controls on live agent actions | Partial | — | ✓ |
| Gateway or proxy in the request path | Partial | — | — |
| Licence and deployment | |||
| Open source | Partial | ✓ | Partial |
| Self-hosting | ✓ | ✓ | Partial |
Langfuse joined ClickHouse in January 2026 and said its roadmap would not change, its licence would not change and self-hosting would stay a first-class option. It has kept shipping since: code evaluators in May, a 1.0 command-line tool in August and evaluator backfills in September.
Dynatrace completed its acquisition of Arize on 1 October 2026. Its announcement says both Phoenix and AX will continue and that Arize capabilities will be integrated into Dynatrace over time. If you already run Dynatrace, that integration may be the deciding factor. If you do not, watch how packaging and pricing change over the next few quarters before committing a large rollout.
This is where the two differ most for a team on a budget. Self-hosted Langfuse includes tracing, datasets and experiments, prompt versioning and online evaluator rules that score live production traffic. You run Postgres, ClickHouse, Redis and object storage; there is no subscription.
Phoenix, Arize's self-hosted tool, includes tracing, datasets, experiments, LLM and code evaluators and prompt versioning. Continuous evaluation of production traffic, monitors, alerting and automated issue detection are Arize AX features, and Phoenix's own documentation points users to AX for them.
Licence matters too. Langfuse is MIT-licensed. Phoenix uses the Elastic License 2.0, which lets you read, modify and self-host the code but forbids offering it to others as a hosted service. For most internal deployments that makes no difference; for a platform team building a service on top of it, check the terms.
Both accept OpenTelemetry traces, so instrumentation is portable between them. Arize maintains OpenInference, a set of OpenTelemetry conventions for LLM and agent spans that several other tools also read. Langfuse accepts OTLP over HTTP (not gRPC) and maps OpenInference-style attributes.
For multi-agent systems, Arize AX draws an agent trajectory graph of which agents call which, set automatically for frameworks such as LangGraph and CrewAI. Langfuse added an agent graph view in July 2026 and groups multi-step runs into sessions.
Both can cost nothing if you self-host: Phoenix and Langfuse are free to run on your own infrastructure, and you pay for the servers and the time to operate them. The difference on self-hosting is scope, not price: self-hosted Langfuse includes online evals, while Phoenix leaves continuous production evals to the paid AX platform.
On the hosted plans, as of 20 October 2026, Langfuse Hobby is free for 50k units a month with 30 days of data, Core is $29 a month for 100k units and then $8 per 100k, and Pro is $199 a month. Arize AX Free covers 25k spans a month with 15-day retention, AX Pro is $50 a month for 50k spans with 30-day retention, and AX Enterprise, the only self-hosted AX option, is priced on request.
A Langfuse unit counts traces, observations and scores, while Arize counts spans, so run a week of your own traffic through each calculator before comparing totals. Check Arize pricing and Langfuse pricing directly; both changed in 2026.
Langfuse does not block or hold anything at run time; its documentation says protection comes from separate libraries and Langfuse evaluates it afterwards. Arize AX documents guards built on the Guardrails AI framework that run in your own code around model calls and can return a default response or re-ask. Neither documents approval workflows or policy checks on an agent's tool calls.
We sell the layer this section describes. Read it with that in mind.
Stated plainly, since Prefactor publishes this page: Prefactor is an agent evaluation platform that scores each step of every run, in development and in production, and can act on a failing score with runtime policies, approval holds and a kill switch for a run, an agent or a team. Its platform is hosted and it has no prompt playground, so if free self-hosting or prompt iteration is the requirement, Langfuse or Phoenix is the better fit. See the full comparison of 15 platforms.
Reviewed against public product and documentation pages on October 20, 2026. If a vendor has changed a feature, product name, or positioning since then, send a correction and we will update it. Numbered source links in the page body point to the ordered sources below.
Prefactor scores each step of every agent run and can hold or stop an action when a score says it should not happen. Start on the free tier: 5,000 spans a month, no credit card required.
Prefactor helps teams observe, evaluate, and improve their AI agents in production, across every framework and provider.