The main Langfuse alternatives in 2026 are Arize Phoenix, LangSmith, Braintrust, Helicone, Opik and Kubit. Phoenix, Opik and Helicone are open source like Langfuse. LangSmith and Braintrust are hosted platforms with self-hosting on enterprise plans. Kubit is different in kind. It labels and scores every span, visually groups agent runs by execution path (which tools ran, in what order) and queries traces natively in your own warehouse, so there is no separate data silo to run.
Key facts
Langfuse is open source (MIT outside its
ee/folder) and free to self-host. Langfuse Cloud starts free with 50k units a month (pricing)Open source alternatives: Arize Phoenix (Elastic License 2.0), Opik (Apache-2.0), Helicone (Apache-2.0)
Hosted alternatives: LangSmith from $39 per seat a month, Braintrust Pro at $249 a month, both with free tiers (LangSmith, Braintrust)
Kubit has no LLM judge cost: Polaris labels and scores every span as part of the price, which is per monthly tracked user rather than per trace (Polaris)
Why do teams look for a Langfuse alternative?
Langfuse is a strong default. It traces and scores every run, it's open source, and you can self-host it for free. Teams usually look further for one of three reasons: they want a managed product with a particular eval workflow, they want a gateway in front of their model calls, or they want to find recurring failures across thousands of runs. Langfuse's own guide does that last part by having someone read a sample of traces and write a note on each one.
How the alternatives compare

Tool | Open source | Self-hosting | Free tier | Paid from | Billed by |
|---|---|---|---|---|---|
Langfuse | Yes, MIT outside | Free | 50k units a month | $29 a month (Core) | Units: traces, observations, scores |
Arize Phoenix / AX | Phoenix, Elastic License 2.0 | Phoenix yes; AX on Enterprise | AX Free: 25k spans a month | $50 a month (AX Pro) | Spans and ingestion |
LangSmith | Not an open source platform | Enterprise | 5k traces, 1 seat | $39 per seat a month | Seats and traces |
Braintrust | Not an open source platform | Enterprise (on-prem) | 1 GB data, 10k scores | $249 a month (Pro) | Data, scores, model credits |
Helicone | Yes, Apache-2.0 | Enterprise (on-prem) | 10k requests | $79 a month (Pro) | Requests and storage |
Opik | Yes, Apache-2.0 | Free (open source) | 25k spans a month | $19 a month (Pro) | Spans |
Kubit | No | Reads traces in your warehouse | Early access | Early access | Monthly tracked users |
Arize Phoenix
Phoenix is Arize's open-source, local-first platform for tracing, evaluation, experimentation and prompt iteration. It's a natural fit if you want to run everything on your own machine first and move to Arize AX, the hosted product, later.
LangSmith
LangSmith is LangChain's observability platform, with traces up to production-wide performance metrics. It suits teams already building on LangChain or LangGraph who want one vendor for the framework and the monitoring.
Braintrust
Braintrust is built around evals: scores, experiments and datasets, with free usage for your whole team on Starter. Its Topics feature has an LLM summarize each trace and clusters the summaries, so you can see what runs were about.
Helicone
Helicone is an AI gateway with observability built in. You point the OpenAI SDK at its gateway URL and every request is logged, which makes it the quickest to add when you mostly want cost, latency and request logs.
Opik
Opik is Comet's open-source platform for tracing every LLM call, tool invocation and agent step, with evaluation and automated prompt optimization. Its Pro cloud plan is the cheapest paid entry on this list.
Kubit
Kubit is Agent Analytics: agent observability with behavior analytics. Polaris, Kubit's in-house small language model, labels and scores every span in under 100ms, with no sampling and no extra LLM cost. It is up to 98% lower cost and ~15x faster than a frontier LLM judge, with comparable accuracy on public agent benchmarks. Polaris marks Missteps, where a call returned OK and the agent still did the wrong thing, and Exceptions, where execution failed. Flows visually groups every run by execution path, so a failure that repeats a thousand times shows up as one path with a count of 1,000. Kubit then joins runs to sessions, users and outcomes. See What is Kubit?.
Can I keep Langfuse and add Kubit?

Yes. The agent plugin adds Kubit's span processor next to the Langfuse processor, so Langfuse keeps receiving every span and Kubit gets a copy. If you self-host Langfuse, Kubit can read its traces in place from your ClickHouse, with nothing changed in your app. Langfuse Cloud traces go through the plugin path. Steps: Use Kubit with Langfuse.
Which Langfuse alternative should I pick?
You want open source you can self-host: Phoenix or Opik
You build on LangChain or LangGraph: LangSmith
Your team lives in evals and experiments: Braintrust
You want a gateway and request logs first: Helicone
You want every run labeled and scored, visually grouped by execution path and joined to user behavior, without moving off Langfuse: Kubit
Agent Analytics in Kubit is in early access. Join the waitlist or book a time with Alex.
Read next: Why we group agent runs by execution path · Why we label every span instead of a sample · What is Kubit?
FAQ
What are the best Langfuse alternatives in 2026?
The main ones are Arize Phoenix, LangSmith, Braintrust, Helicone, Opik and Kubit. Phoenix, Opik and Helicone are open source. LangSmith and Braintrust are hosted platforms with self-hosting on enterprise plans. Kubit labels and scores every span, visually groups runs by execution path and queries traces in your own warehouse.
Is there an open source alternative to Langfuse?
Yes. Arize Phoenix is source-available under the Elastic License 2.0, and Opik and Helicone are open source under Apache-2.0. Langfuse itself is open source under MIT outside its enterprise folder, and free to self-host.
Can I use Kubit with Langfuse?
Yes. Kubit's agent plugin adds its span processor next to the Langfuse processor, so Langfuse keeps receiving every span. Self-hosted Langfuse traces in ClickHouse can be read in place with no change to your app.
How is Langfuse priced?
Self-hosting is free. Langfuse Cloud has a free Hobby plan with 50k units a month, then Core at $29 a month and Pro at $199 a month, with extra usage at $8 per 100k units. A unit is any trace, observation or score.
Keep reading

What is Kubit? Agent Analytics, from trace to user

Misstep vs Exception: why your agent fails when its traces are green

Braintrust alternatives for agent analytics in 2026

Agent analytics on traces already in Snowflake, Databricks or BigQuery

Should you build your own agent analytics?

