Langfuse pricing in 2026: tiers, self-hosting, and the ClickHouse factor
By Coverge Team
Langfuse is an open-source LLM observability platform for tracing, evaluating, and monitoring AI applications. Since ClickHouse acquired Langfuse in January 2026, the product has kept its MIT license and added no new pricing gates — but the acquisition reshapes how you should think about self-hosting and long-term vendor risk.
This guide covers what each cloud tier includes, how self-hosting economics work, and where Langfuse fits against alternatives like LangSmith and Braintrust.
Langfuse cloud pricing tiers
Langfuse Cloud offers four tiers. All paid tiers include unlimited users — there is no per-seat charge at any level.
| Hobby (Free) | Core ($29/mo) | Pro ($199/mo) | Enterprise ($2,499/mo) | |
|---|---|---|---|---|
| Included units/mo | 50,000 | 100,000 | 100,000 | 100,000 |
| Data retention | 30 days | 90 days | 3 years | 3 years |
| Users | 2 | Unlimited | Unlimited | Unlimited |
| Overage rate | Hard cap (no overage) | $8/100K units (graduated) | $8/100K units (graduated) | $8/100K units (graduated) |
| Support | Community (GitHub/Discord) | In-app, 48hr SLO | Prioritized in-app | Dedicated engineer |
| Compliance | — | — | SOC 2 Type II, ISO 27001, HIPAA | SOC 2 Type II, ISO 27001, HIPAA |
| Ingestion throughput | Standard | Standard | 20K req/min | Custom |
| Custom volume pricing | No | No | No | Yes (yearly commitment) |
Hobby tier: enough to evaluate, not enough to ship
The free Hobby tier gives you 50,000 units per month with 30-day data retention and two user seats. That is 10x the free volume LangSmith offers (5,000 traces), which makes Langfuse a more realistic option for evaluating observability before committing budget.
The 50,000-unit cap is a hard limit — there is no overage billing, so your pipeline stops reporting when you hit it. For a simple LLM application with one model call per request, 50,000 units translates to roughly 25,000 user requests (each request generates a trace plus an observation). For a RAG pipeline with retrieval, reranking, and generation steps, that number drops to around 8,000-10,000 requests.
The two-user limit is the real constraint for teams. If more than two people need to look at traces — and they will once you are past prototyping — you need Core.
Core tier: production entry point
Core at $29 per month includes 100,000 units, 90-day retention, unlimited users, and in-app support with a 48-hour response SLO. This is where most small production teams start.
The unlimited users matter. A five-person team on Langfuse Core pays $29 per month total. The same team on LangSmith Plus pays $195 per month ($39 per seat times five) before counting trace overages. That is a 6.7x price difference at the seat level alone.
Overage pricing is usage-based and graduated:
| Monthly usage band | Rate per 100K units |
|---|---|
| 100K – 1M | $8.00 |
| 1M – 10M | $7.00 |
| 10M – 50M | $6.50 |
| 50M+ | $6.00 |
The graduated structure means your effective per-unit cost drops automatically as volume increases — no sales negotiation required until you want Enterprise-level custom pricing.
Here is how Core tier costs scale for different traffic profiles:
| Monthly units | Base cost | Overage | Total monthly |
|---|---|---|---|
| 100,000 | $29 | $0 | $29 |
| 500,000 | $29 | $32 | $61 |
| 1,000,000 | $29 | $72 | $101 |
| 5,000,000 | $29 | $352 | $381 |
| 10,000,000 | $29 | $702 | $731 |
At 5 million units per month, you are paying $381 total on Langfuse Core. A comparable LangSmith setup with a 10-person team and similar trace volume would run $4,000-6,000 per month depending on retention settings.
Pro tier: compliance and throughput
Pro at $199 per month adds SOC 2 Type II, ISO 27001, and HIPAA compliance certifications. The included unit volume is the same 100,000 as Core, but you get 3-year data retention (versus 90 days) and higher ingestion throughput at 20,000 requests per minute.
Pro makes sense when your compliance team requires certified infrastructure. The $170 premium over Core buys you 3-year retention and the compliance paperwork — both of which are expensive to replicate with self-hosting.
An optional Teams Add-on at $300 per month provides additional collaboration features for organizations managing multiple projects.
Enterprise tier: dedicated support and volume pricing
Enterprise at $2,499 per month includes everything in Pro plus a dedicated support engineer and the ability to negotiate custom volume pricing on a yearly commitment. The base included units are still 100,000, but enterprise customers typically negotiate significantly lower per-unit rates.
Teams move to Enterprise when they need custom volume pricing (typically above 50 million units per month), a dedicated support contact, or contractual SLAs beyond what Pro offers.
How Langfuse counts billable units
Understanding Langfuse's billing model requires knowing what counts as a "unit." A billable unit is the sum of three event types:
- Traces: one per user request or pipeline invocation
- Observations: one per LLM call, span, or event within a trace
- Scores: one per evaluation score (including automated LLM-as-a-judge scores, annotation queue scores, and experiment scores)
A single user request through a RAG pipeline might generate: 1 trace + 4 observations (retrieval, reranking, generation, safety check) + 1 evaluation score = 6 billable units.
This means the ratio between "user requests" and "billable units" depends heavily on your pipeline complexity. Simple single-call applications run close to 2 units per request. Complex agent workflows with tool calls, reflection loops, and automated evaluation can generate 15-20 units per request.
Before committing to a tier, instrument your pipeline and measure your actual units-per-request ratio. Multiply by your expected monthly request volume to estimate your tier requirements.
Self-hosting: the MIT advantage
Langfuse's core product is MIT-licensed. You can self-host the entire tracing, evaluation, prompt management, and monitoring stack with no license cost and no event limits. Deployment templates are available for Docker, Kubernetes, AWS, Azure, and GCP.
In June 2025, Langfuse open-sourced all remaining cloud-only product features under MIT, including managed LLM-as-a-judge evaluations, annotation queues, and prompt experiments. The only features that still require an enterprise license key for self-hosted deployments are:
- SCIM (directory sync)
- Audit logs
- Data retention policies
- Enterprise support SLAs
Everything else — tracing, dashboards, prompt management, datasets, the playground, evaluations — runs with no license key, no usage metering, and no phone-home.
Self-hosting cost reality
"Free" self-hosting is not free. Langfuse v3 migrated its data layer to ClickHouse, so a self-hosted deployment requires:
- ClickHouse cluster for trace storage and analytics — this is the primary cost driver. Expect $200-800/month on cloud infrastructure depending on trace volume and retention requirements.
- PostgreSQL database for metadata and configuration — relatively cheap at $20-50/month.
- Application servers running the Langfuse container — $50-150/month depending on throughput.
- Operations overhead — someone needs to monitor, upgrade, and scale the infrastructure.
For a mid-size team generating 10 million events per month with 90-day retention, self-hosting infrastructure typically costs $400-1,000 per month. That is comparable to or cheaper than Langfuse Cloud at the same volume ($680 on Core), but you are trading money for operations complexity.
Self-hosting makes clear financial sense in two scenarios:
- High volume (50M+ events/month): cloud overage costs at $6/100K units add up to $3,000+ per month. Self-hosted ClickHouse scales more cost-effectively at this volume.
- Data sovereignty requirements: some regulated industries cannot send trace data to a third-party cloud, regardless of compliance certifications.
For teams under 5 million events per month without data residency requirements, Langfuse Cloud is almost always the better deal once you account for operations time.
The ClickHouse acquisition: what changed
ClickHouse acquired Langfuse on January 16, 2026, alongside a $400 million Series D funding round. The acquisition formalized a technical dependency that already existed — Langfuse v3 had migrated its core data layer to ClickHouse because PostgreSQL could not handle the combined high-throughput ingestion and fast analytical reads that production LLM observability requires.
What changed:
- Pricing: nothing. The same tiers, same rates, same structure.
- MIT license: unchanged. The acquisition announcement explicitly stated no planned licensing changes.
- Self-hosting: still fully supported. ClickHouse now offers bundled infrastructure plans for enterprise self-hosted customers (Langfuse fees stack on top of ClickHouse Cloud/BYOC fees), but the MIT self-hosting path is unaffected.
- Long-term direction: Langfuse is now part of ClickHouse's broader data platform strategy. This likely means deeper ClickHouse integration and potentially faster feature development funded by ClickHouse's resources.
The acquisition reduces one risk (Langfuse's standalone financial viability) and introduces another (dependency on ClickHouse's strategic priorities). For most teams, the net effect is positive — more resources behind the product with no immediate cost or licensing changes.
How Langfuse pricing compares to alternatives
| Tool | Free tier | Paid starting at | Per-seat pricing | Open source | Self-host option |
|---|---|---|---|---|---|
| Langfuse | 50K units, 2 users | $29/mo | No | Yes (MIT) | Yes |
| LangSmith | 5K traces, 1 seat | $39/seat/mo | Yes ($39/seat) | No | No |
| Braintrust | 1M spans | $249/mo | No (unlimited) | No | No |
| Helicone | 100K requests | $30/mo | No | Yes | Yes |
| Arize Phoenix | Unlimited (self-hosted) | Enterprise only | No | Yes | Yes |
Langfuse vs LangSmith pricing
This is the comparison most teams make. The two platforms count events differently — Langfuse counts traces, observations, and scores as separate billable units, while LangSmith counts at the trace level (one trace per request, regardless of internal spans). A pipeline generating 1 million Langfuse units typically corresponds to roughly 200,000-300,000 LangSmith traces, depending on pipeline complexity.
For a 7-person team generating roughly 250,000 user requests per month (about 1 million Langfuse units or 250,000 LangSmith traces):
- Langfuse Core: ~$101/month ($29 base + $72 overage for 900K units beyond included)
- LangSmith Plus (14-day retention): ~$873/month ($273 seats + ~$600 overage for ~240K traces beyond included)
- LangSmith Plus (400-day retention): ~$1,473/month ($273 seats + ~$1,200 overage)
Langfuse is 9-15x cheaper for this team profile. The gap comes from two structural differences: no per-seat charge and lower per-event overage rates.
LangSmith's advantages are tighter LangChain integration and a more polished debugging UI for LangChain-specific workflows. If your entire stack is LangChain and LangGraph, the integration convenience may justify the premium. For everyone else, the price difference is hard to ignore. See our full Langfuse vs LangSmith comparison for feature-by-feature analysis.
Langfuse vs Braintrust pricing
Braintrust offers a more generous free tier (1 million spans versus 50,000 units) and also charges no per-seat fees. At the paid level, Braintrust's Pro tier starts at $249 per month — higher than Langfuse Core's $29 but lower than Langfuse Enterprise's $2,499.
The tools have different strengths. Langfuse focuses on tracing and observability with strong prompt management. Braintrust focuses on evaluation and datasets with a sophisticated scoring framework. Many teams use both — Langfuse for production tracing and Braintrust for eval pipelines. See our Braintrust comparison for details.
What Langfuse does not cover
Langfuse is an observability and evaluation platform. It excels at tracing LLM calls, managing prompts, and running evaluations. It does not address:
Deployment governance. Langfuse can track evaluation scores and flag regressions, but it does not enforce deployment gates. There is no built-in mechanism to block a deployment when eval scores drop below a threshold. The decision to ship remains manual.
AI pipeline orchestration. Langfuse monitors pipelines but does not run them. It traces what happened; it does not control what happens next. If you need automated pipeline execution with built-in governance, you are looking at a different layer of the LLMOps stack.
Automated rollback. When production metrics degrade, Langfuse surfaces the problem through dashboards and alerts. Rolling back to a known-good configuration requires your own deployment infrastructure.
Compliance audit trails. Langfuse logs trace and evaluation data, but it does not produce structured proof bundles — the combination of eval results, approval records, and deployment metadata that compliance teams need. For regulated AI deployments requiring AI governance controls, this means maintaining documentation outside the platform — see our AI audit trail guide for what that entails.
These gaps reflect Langfuse's design scope, not a failing. For the governance and deployment lifecycle layer, tools like Coverge address what observability platforms leave open.
When Langfuse is the right choice
Langfuse is worth adopting when:
- Budget matters and your team is growing. No per-seat pricing means adding engineers, PMs, and data scientists costs nothing extra. This is Langfuse's single biggest advantage over LangSmith.
- You want optionality on self-hosting. The MIT license means you can start on Langfuse Cloud and move to self-hosted later without rewriting your instrumentation.
- You use multiple frameworks or no framework. Langfuse's SDK works with LangChain, LlamaIndex, OpenAI, Anthropic, and custom code. You are not locked into one ecosystem.
- Your trace volume is high. Graduated pricing at $6-8 per 100K units is significantly cheaper than LangSmith's $2.50-5.00 per 1K traces at scale.
Langfuse is harder to justify when:
- You are deeply invested in LangChain. LangSmith's integration with LangChain and LangGraph is tighter and the trace visualization is purpose-built for LangChain's abstractions.
- You need compliance certifications at low spend. SOC 2 and HIPAA require the Pro tier at $199/month — the free and Core tiers do not include compliance certifications.
- You want a one-stop evaluation platform. Langfuse's evaluation features are solid but not as deep as Braintrust's purpose-built eval framework.
FAQ
Is Langfuse free?
Langfuse has a free Hobby tier with 50,000 units per month, 30-day data retention, and two user seats. The free tier has a hard cap — there is no overage billing, so tracing stops when you hit the limit. Langfuse is also MIT-licensed and can be self-hosted at no license cost with unlimited events and unlimited users.
How much does Langfuse cost per month?
Langfuse Cloud starts at $29 per month on the Core tier with 100,000 included units and unlimited users. The Pro tier at $199 per month adds compliance certifications and 3-year retention. Enterprise starts at $2,499 per month with dedicated support and custom volume pricing. All tiers charge $6-8 per 100K units for overage on a graduated scale.
Is Langfuse cheaper than LangSmith?
For most team sizes and traffic levels, significantly. A 7-person team with 1 million monthly events pays roughly $98 on Langfuse Core versus $1,500-2,800 on LangSmith Plus depending on retention settings. The structural difference is that Langfuse charges no per-seat fees and has lower per-event overage rates.
Can I self-host Langfuse for free?
Yes. Langfuse's core product is MIT-licensed with no usage limits or license keys required for self-hosting. You pay only for infrastructure (ClickHouse, PostgreSQL, application servers). The only features requiring a commercial license for self-hosted deployments are SCIM, audit logs, data retention policies, and enterprise support SLAs.
What happened with the ClickHouse acquisition?
ClickHouse acquired Langfuse on January 16, 2026, alongside a $400 million Series D funding round. No pricing, licensing, or self-hosting changes resulted from the acquisition. Langfuse remains MIT-licensed and operates with the same cloud pricing tiers. The acquisition formalized an existing technical relationship — Langfuse v3 already ran on ClickHouse.
How does Langfuse count billable units?
A billable unit is the sum of traces, observations, and scores. Each trace (one per request) counts as one unit. Each observation (LLM call, span, or event within a trace) counts as one unit. Each score (evaluation result) counts as one unit. A typical RAG pipeline request generates 4-8 billable units depending on complexity.