Langfuse pricing

About

Langfuse is an open-source LLM engineering platform — observability and tracing, evaluations, prompt management, metrics, and a playground — built during Y Combinator’s W23 batch right after GPT-4 launched. It raised a $4M seed from Lightspeed Venture Partners, La Famiglia, and Y Combinator, and grew into one of the most popular LLMOps tools on the market: roughly 28.7k GitHub stars, 300+ contributors, and 40,000+ builders by 2026.

On January 16, 2026, ClickHouse acquired Langfuse alongside its own $400M Series D at a $15B valuation. The deal formalized a dependency that already existed — Langfuse v3 had migrated its core data layer to ClickHouse because PostgreSQL couldn’t handle simultaneous high-throughput trace ingestion and fast analytical reads. ClickHouse stated there are no planned changes to Langfuse pricing, licensing, or self-hosting. The product remains MIT-licensed and self-hostable.

For the most current information, visit Langfuse.

Pricing summary : How Langfuse’s pricing model works

Langfuse runs an open-core model with two front doors. You can self-host the full platform for free under MIT with unlimited usage, or use Langfuse Cloud, which is metered on units. A unit is one trace, one observation (an LLM call, span, or event inside a trace), or one score (an evaluation result), and they are summed — so a single request that fires three LLM calls and produces two scores counts as six units.

Cloud has four tiers — Hobby (free), Core ($29/mo), Pro ($199/mo), Enterprise ($2,499/mo) — each bundling included units (50k or 100k), with the same graduated overage of $8 per 100k units, falling to $6 at scale. Notably, paid tiers charge no per-seat fee (unlimited users from Core up); seats only re-enter via the optional $300/mo Teams add-on for SSO/RBAC.

What makes this different: the value metric is ingestion volume, not seats or per-trace. Because observations and auto-generated eval scores each count, agentic apps with deep call trees consume units far faster than tools that bill one unit per request — a deliberate alignment of price with the platform’s actual storage/compute cost (ClickHouse rows).

Pricing by product

Sales motions across products: open-source (self-host), self-serve credit-card checkout (Hobby/Core/Pro), and sales-led (Enterprise yearly commitments, AWS Marketplace, invoicing).

Hidden costs : What Langfuse users actually pay

The headline tier price is rarely the real bill — unit overage is. The “unit = trace + observation + score” definition is the single biggest surprise: an agentic pipeline can multiply a request into 5–10 units, and auto-generated eval scores (LLM-as-a-judge, annotation queues, experiments) add more on top. A team modeling its bill on request count will undershoot badly.

For a Core team ingesting ~1M units/month, the math is roughly:

Other line items to watch: the Teams add-on ($300/mo) is the only way to get SSO/RBAC/project-level controls below Enterprise; HIPAA/SOC2 reports start at Pro ($199); and self-hosted Enterprise is additive to your ClickHouse commercial plan, so the database cost is a separate bill.

Want to estimate your own Langfuse bill? Use the Langfuse pricing calculator to model your costs based on usage patterns.

Pricing evolution : Langfuse pricing history and changes

Cadence

Tracked range: 2023–present. Wayback CDX access returned 403 at capture time (2026-06-09), so snapshot images are not attached; evolution is reconstructed from official posts and community teardowns.

Notable changes

• 2023 — Launched open source during YC W23; free self-host plus metered Cloud.
• 2024 — v3 migrated the data layer to ClickHouse to handle ingestion + analytics at scale.
• 2025 — Cloud settled on the unit model (traces + observations + scores) with graduated $8→$6 per-100k tiers.
• 2026-01-16 — Acquired by ClickHouse at a $15B valuation; explicitly no change to pricing, licensing, or self-hosting.

What’s unique : Langfuse’s distinctive pricing mechanics

1. The “unit” value metric. Billing on the sum of traces, observations, and scores ties price directly to ingested data volume (ClickHouse rows). It scales with how instrumented your app is, not how many people look at it — a fundamentally different lever than seat-based observability tools.

2. Seat-free paid tiers. Core, Pro, and Enterprise all include unlimited users. Seats only reappear as the optional $300/mo Teams add-on for SSO/RBAC. This removes the per-engineer tax that deters broad adoption of observability inside a company.

3. A genuinely free, unlimited self-host path. Unlike “open-core” tools that cripple the free build, the MIT self-host has no usage or seat caps and includes all core features. Monetization comes from Cloud convenience and enterprise governance (SSO/audit/SLA), not from gating the core.

Strengths & weaknesses

Billing UX : Langfuse billing controls and transparency

• Billing controls — Self-serve subscription management for Hobby/Core/Pro; usage-fee alerts can be configured. Enterprise yearly commitments and custom volume pricing go through sales.
• Usage visibility — An interactive pricing calculator on the pricing page breaks down graduated tiers live; the app surfaces unit consumption so teams can see overage building. Docs explain the billable-unit definition in detail.
• Payment options — Credit card for Core/Pro; Enterprise adds invoicing and AWS Marketplace billing. Standard T&Cs and DPA on self-serve tiers; redline-able contracts at Enterprise.

Strategic wins : Why Langfuse’s pricing decisions worked

1. Cost-aligned value metric

By metering on units (ClickHouse rows), Langfuse made revenue track infrastructure cost almost 1:1. Heavy users pay more because they genuinely consume more storage and compute — a clean, defensible model that scales without renegotiation. See choosing the right usage metric for why this matters.

2. Open source as the top of funnel

The free MIT self-host built ~28.7k stars and 40,000+ builders, seeding adoption inside thousands of companies before any sales conversation. Cloud convenience and enterprise governance then convert the highest-intent users. Related: how AI companies structure pricing and usage-based pricing fundamentals.

3. Removing the per-seat tax

Unlimited users on paid plans let observability spread across an entire engineering org without a budgeting fight — the opposite of seat-metered tools. Pricing follows usage, which is exactly what an instrumentation product wants. See outcome-based pricing trends.

Areas to improve : Gaps in Langfuse’s pricing approach

1. Unit-count predictability

The trace + observation + score formula is precise but unintuitive. Teams routinely underestimate bills because agentic call trees and auto-generated eval scores inflate unit counts 3–10x over naive per-request math. Clearer in-product forecasting would reduce bill shock.

2. Governance gated high

SSO, RBAC, and project-level access live behind the $300/mo Teams add-on or Enterprise. For security-conscious mid-market teams, that’s a steep step up from the $29 Core tier just to get SSO.

3. The ClickHouse-additive self-host bill

Self-hosted Enterprise pricing is “additive to your ClickHouse commercial plan,” which is honest but makes total cost of ownership opaque — buyers must price two vendors (now one parent) to model the real number.

Key takeaways

1. Price on the metric that mirrors your cost. Units = ClickHouse rows keeps Langfuse’s revenue and COGS in lockstep, so scale never breaks the model.
2. A real free self-host is a moat, not a leak. Unlimited MIT self-host drove adoption that paid acquisition-grade attention — ClickHouse bought the distribution as much as the product.
3. Drop the seat tax for instrumentation tools. Unlimited users on every paid tier let observability spread org-wide; seats only return as an enterprise-governance add-on.
4. Transparency includes the gotchas. Langfuse publishes graduated tiers and documents the unit definition — but the multiplicative unit math still surprises buyers, so forecasting UX matters as much as a clear price page.
5. Usage metrics survive M&A. The ClickHouse acquisition changed ownership, not pricing — a sign the unit model was already structurally sound.

UBP implications

1. Choose a value metric tied to infrastructure cost when your product’s marginal cost is storage/compute — it makes graduated volume discounts honest and renewal-proof.
2. Composite usage units (trace + observation + score) need first-class forecasting tooling, or you trade pricing elegance for buyer trust. Build the calculator and in-app meter before you ship the metric.
3. Open-core works when the free tier is uncrippled: monetize convenience (hosting) and governance (SSO/audit/SLA), not the core capability, and let usage-based Cloud pricing capture the value at scale.

Sources

• Langfuse Cloud pricing page (accessed 2026-06-09)
• Langfuse self-host pricing (accessed 2026-06-09)
• Langfuse billable-units documentation (accessed 2026-06-09)
• ClickHouse: acquiring Langfuse + $400M Series D (accessed 2026-06-09)
• Langfuse: joining ClickHouse (accessed 2026-06-09)
• Langfuse on GitHub (YC W23) (accessed 2026-06-09)

Bottom line

Langfuse prices the way an instrumentation product should: free and unlimited if you self-host the MIT core, and metered on units — the sum of traces, observations, and scores — if you use Langfuse Cloud. Seat-free paid tiers ($29 Core, $199 Pro, $2,499 Enterprise) with graduated $8→$6 per-100k overage keep revenue aligned to ClickHouse-row cost. The January 2026 ClickHouse acquisition (at a $15B valuation) left pricing untouched — proof the unit model was already sound. The one real friction is predicting unit consumption for agentic apps, where each request can fan out into many billable units.

Want to compare Langfuse against other AI infrastructure companies? Browse the pricing blueprint.