Arize Phoenix vs LangSmith

Phoenix is Arize's open-source, self-hosted tool. It is not Arize AX, Arize's paid hosted platform. This page compares Phoenix with LangSmith.

Most vendor "Arize vs LangSmith" pages compare Arize AX, not Phoenix. Arize's own Phoenix docs have a separate Phoenix-vs-LangSmith FAQ, listed below.

Pick Arize Phoenix if you want to run tracing and evals on your own machine or servers at no licence cost, and you are fine with the Elastic License 2.0 and running Phoenix yourself.

Pick LangSmith if you want a hosted service run for you, you build on LangChain or LangGraph, and you are happy to pay per seat and per trace once you pass 5k traces a month.

If you need Phoenix-style tooling but hosted, look at Arize AX in the table below. It is a different product from Phoenix.

Updated · Evals Review · No affiliate links on this page.

How we rank: We don't pick a winner: the verdict says when each option fits, based on the sourced facts in the table. Every entry gets the same fields, taken from vendor pricing pages, docs and official repositories on the checked date, with a source link; anything we couldn't confirm says “not verified”. Commission and sponsorship don't affect order or what we say. We show no star ratings or user scores. How we review

The facts side by side

Arize PhoenixLangSmith
Open source or hostedOpen source; you host it [source, checked 2026-11-01]Hosted service; the SDK is open source (MIT) [source, checked 2026-11-01]
Self-hostYes, free (pip, Docker or Kubernetes) [source, checked 2026-11-01]Enterprise plan only (self-hosted or hybrid) [source, checked 2026-11-01]
LicenceElastic License 2.0 (source-available, not OSI-approved) [source, checked 2026-11-01]Platform: proprietary. SDK: MIT [source, checked 2026-11-01]
OwnerArize AI, part of Dynatrace since 1 Oct 2026 [source, checked 2026-11-01]LangChain [source, checked 2026-11-01]
Price (checked 1 Nov 2026)Phoenix: free. Hosted alternative Arize AX: Free (25k spans/mo), Pro $50/mo (50k spans), Enterprise custom [source, checked 2026-11-01]Developer $0 (1 seat, 5k base traces/mo), Plus $39/seat/mo (10k base traces/mo), then pay as you go; Enterprise custom [source, checked 2026-11-01]
IntegrationsOpenTelemetry/OpenInference; OpenAI, Anthropic, Bedrock, LiteLLM, LangChain, LangGraph, LlamaIndex, CrewAI, DSPy, and others on its integrations page [source, checked 2026-11-01]OpenTelemetry endpoint; OpenAI, Anthropic, Bedrock, LiteLLM, LangChain, LangGraph, CrewAI, AutoGen, Google ADK, and others on its integrations page [source, checked 2026-11-01]

Prices as of November 2026 (checked 2026-11-01 on both vendors' own pricing pages); check the vendor site before you buy. Prices are list prices in USD; taxes and usage fees can apply.

LangSmith also bills usage in LangChain Standard Units (1 LSU = $1) for Deployment, Fleet, Engine and Sandboxes; see its pricing page. Base traces keep 14 days; extended traces 180 days.

Ownership note

  • 1 Oct 2026: Dynatrace says its acquisition of Arize is complete. It also says Phoenix remains an open-source project and Arize AX remains available. [source, checked 2026-11-01]

What we ran ourselves (2026-11-01)

Same 5-question dataset in both tools. The app returns fixed answers (4 right, 1 wrong on purpose), so no model is called and nothing is paid for. The scorer checks whether the reference answer appears in the output. Phoenix ran as a local server on our test machine. LangSmith ran with its Python SDK's `evaluate(..., upload_results=False)` and no API key, so nothing was sent to LangSmith.

Arize PhoenixLangSmith
Version testedarize-phoenix 21.0.0, arize-phoenix-client 3.6.0langsmith (Python SDK) 0.14.16
How it ranLocal server + `client.experiments.run_experiment` with a code evaluator`langsmith.evaluate` with a code evaluator, results kept local
Scores (5 rows)1, 1, 0, 1, 11, 1, 0, 1, 1
Setup notesNo account. Server start plus run took about 6 seconds. Results visible in the local UINo account needed for this offline mode. No UI without a LangSmith account

Not tested:

  • Hosted LangSmith (tracing UI, datasets, online evals): it needs a sign-up, so we did not use it.
  • Arize AX: hosted, needs a sign-up.
  • LLM-as-judge evaluators in either tool: we kept the test deterministic and free.
  • Scale, latency under load, or retention limits.

Both tools agreed on every row, as expected with a deterministic scorer. This shows that the basic eval loop works locally. It does not rank the tools.

Where a vendor's own comparison favours itself

ClaimWho says itHow to read it
Arize vs. LangSmithArize (vendor of one side) [source, checked 2026-11-01]Arize's page compares mostly Arize AX, its paid product. Read its Phoenix rows on their own, and check prices on LangChain's pricing page.
LangSmith vs ArizeLangChain (vendor of the other side) [source, checked 2026-11-01]LangChain's page compares LangSmith against Arize as a whole. Its hosting and deployment points mostly apply to AX, not to self-hosted Phoenix.
Open-source LangSmith alternativeArize Phoenix docs [source, checked 2026-11-01]This is a vendor FAQ that recommends Phoenix. Check the licence: Phoenix is under ELv2, not an OSI licence.

FAQ

Is Arize Phoenix the same as Arize AX?

No. Phoenix is the open-source tool you host yourself. Arize AX is Arize's paid hosted platform, with Free, Pro ($50/mo) and Enterprise plans as of 1 Nov 2026.

Can I self-host LangSmith for free?

No. LangChain's pricing page lists self-hosted and hybrid deployment only on the Enterprise plan.

Is Phoenix open source?

Its code is public under the Elastic License 2.0. That licence is source-available, not OSI-approved, and it limits offering Phoenix as a managed service.

Did you test hosted LangSmith?

No. We ran only the LangSmith Python SDK in offline mode, with no account. See "What we ran ourselves".

Related

Sources