LangSmith vs Langfuse for teams running production agents
LangSmith is LangChain's hosted tracing and evals platform, self-hostable only on its Enterprise plan, while Langfuse is an MIT-licensed open-source platform, owned by ClickHouse since January 2026, that any team can self-host free, so the choice turns on where traces must live and how much of the stack already runs on LangChain.
Checked
LangSmith and Langfuse cover the same ground for a team running agents in production: traces of every model and tool call, token and cost tracking, datasets, offline and online evals, and prompt management. Both accept OpenTelemetry traces, and both appear in Anthropic's January 2026 guide to agent evals.
The differences are in ownership, licence and hosting. LangSmith is a proprietary platform from LangChain with MIT-licensed SDKs. Langfuse is open source under MIT for its core, and since 16 January 2026 it belongs to ClickHouse, which said there are no planned changes to licensing.
Ownership and licence in 2026
LangSmith
Built by LangChain. The platform is closed source and the client SDKs are MIT. LangSmith works with any framework, with native support for LangChain and LangGraph.
Langfuse
Acquired by ClickHouse on 16 January 2026. Langfuse stays open source and self-hostable. The core is MIT and the ee/ directories carry a separate commercial licence.
Pricing and self-hosting
LangSmith is priced per seat plus traces: Developer is $0 for one seat and 5k base traces a month, Plus is $39 a seat a month with 10k base traces, and Enterprise is custom. Base traces are kept 14 days, and extended retention of 180 days costs extra. Self-hosted and hybrid deployments are Enterprise only.
Langfuse Cloud is priced by units: Hobby is free for 50k units a month with 30 days of data access and 2 users, Core is $29 a month for 100k units and 90 days, Pro is $199 a month for 100k units and 3 years, and Enterprise is $2,499 a month. Overage starts at $8 per 100k units. The self-hosted open-source edition is free without limits on core features; Self-Hosted Enterprise and Government (FIPS 140-3, air-gapped) editions are custom-priced on top of a ClickHouse commercial plan.
Tracing, OpenTelemetry and cost tracking
OpenTelemetry
LangSmith accepts OTLP over HTTP at its /otel endpoint. Langfuse accepts OTLP over HTTP (JSON or protobuf) at /api/public/otel, does not yet accept gRPC, and its Python and JS/TS SDKs are a thin layer on the official OpenTelemetry client. The OpenTelemetry GenAI semantic conventions both build on are still at Development status.
Cost tracking
LangSmith computes cost per trace from its model pricing table, or from costs sent in usage metadata for tool calls and non-linear pricing. Langfuse computes cost per generation from model price definitions, or uses costs ingested from the SDK, which take priority.
Evals
LangSmith
Offline and online evals with datasets. LangSmith Engine, in beta since May 2026, scans production traces every 6 hours by default, clusters failures into issues, proposes fixes and generates evaluators and dataset examples. Engine is billed in LangChain Compute Units.
Langfuse
LLM-as-a-judge and code evaluators on live traces and on datasets, experiments that compare prompt, model or code changes, a GitHub Action that fails a pull request on a regression, annotation queues for human review and score analytics to check a judge against human labels.
How to decide
A team whose agents run on LangChain or LangGraph, and that is comfortable with a hosted platform, gets one vendor across runtime, tracing and evals with LangSmith. A team that must keep traces on its own infrastructure, or wants to avoid per-seat pricing, can run Langfuse's open-source edition without a contract. Teams migrating off the OpenAI Evals platform, which went read-only on 31 October 2026 and shuts down on 30 November 2026, can rebuild dataset evals in either.
Motiontic's engineers are vendor-neutral on tracing and eval tools. They pick a backend against the client's hosting, data-residency and budget constraints, keep instrumentation portable where the stack allows, build the eval suite from the client's own production failures, and leave the traces, datasets and evals with the client's team.
LangSmith and Langfuse compared for production agents. Prices in USD, as of 1 November 2026
| LangSmith | Langfuse | |
|---|---|---|
| Owner | LangChain | ClickHouse, since 16 January 2026 |
| Licence | Proprietary platform; SDKs MIT | Core MIT; ee/ directories under a commercial licence |
| Self-hosting | Enterprise plan only (self-hosted or hybrid) | Open-source edition free; Enterprise and Government editions custom-priced |
| Cloud entry pricing | Developer $0 (1 seat, 5k base traces); Plus $39 a seat a month (10k) | Hobby $0 (50k units, 2 users); Core $29 a month; Pro $199 a month |
| Data retention | 14 days base; 180 days extended, for a fee | 30 days on Hobby, 90 on Core, 3 years on Pro |
| OpenTelemetry | Accepts OTLP over HTTP | Accepts OTLP over HTTP; SDKs built on the OTel client; no gRPC yet |
| Evals | Offline and online, datasets, Engine (beta) generates evaluators from failures | LLM-as-a-judge and code evaluators, online and offline, experiments, CI action, annotation queues |
| Cost tracking | Per trace, from a model pricing table or sent costs | Per generation, from model price definitions or ingested costs |
Common questions
Is Langfuse still open source after the ClickHouse acquisition?
Yes. ClickHouse acquired Langfuse on 16 January 2026, and Langfuse said it stays open source and self-hostable with no planned changes to licensing. The core remains MIT, with enterprise features in ee/ directories under a commercial licence.
Can LangSmith be self-hosted?
Only on the Enterprise plan, which lists self-hosted and hybrid deployments. Developer and Plus plans are hosted by LangChain.
Is LangSmith or Langfuse cheaper?
Both have a free tier. LangSmith Plus is $39 a seat a month with 10k base traces, while Langfuse Cloud Core is $29 a month for 100k units regardless of seats, and Langfuse's open-source edition is free to self-host. The real cost depends on trace volume, team size, retention needs and the cost of running infrastructure for a self-hosted install.
Do LangSmith and Langfuse support OpenTelemetry?
Both accept OTLP traces over HTTP. Langfuse's Python and JS/TS SDKs are built on the OpenTelemetry client. The OpenTelemetry GenAI semantic conventions are still at Development status, so attribute names can change.
Can LangSmith or Langfuse replace the OpenAI Evals platform?
OpenAI's named migration path is Promptfoo, and the Evals platform shuts down on 30 November 2026. Teams that also want traces and production scoring in one place can move dataset evals to LangSmith or Langfuse, both of which run offline and online evals.
Related
AI agent development
Agents that do real work for users, with the evals, traces and guardrails to keep them in production.
LangSmith alternatives
Eight tracing and eval platforms compared on licence, hosting, strengths and the teams each one fits.
Langfuse alternatives
Seven tracing and eval platforms compared on licence, hosting, strengths and the teams each one fits.
Braintrust vs LangSmith
Two hosted platforms for agent traces and evals, one built around experiments, one around the LangChain stack.
Agent observability
Step-by-step traces of what an agent did, why, and what it cost.
Sources
- LangChain: LangSmith pricing
- LangSmith docs: Trace with OpenTelemetry
- LangSmith docs: Cost tracking
- LangSmith docs: Engine (beta)
- Langfuse: Joining ClickHouse (2026-01-16)
- Langfuse: Cloud pricing
- Langfuse: Self-host pricing
- Langfuse docs: OpenTelemetry
- Langfuse docs: Token and cost tracking
- Langfuse docs: Evaluation overview
- OpenAI: API deprecations
Running agents that have to hold up?
Describe the agent, the systems it touches and what it must never get wrong. The first call is with a lab engineer, who will say plainly whether it is ready for production, and whether Motiontic is the right team.