Langfuse Pricing Explained (2026) - What You Actually Pay

Langfuse is the open-source observability default, and its pricing is genuinely fair. But “fair” is not the same as “obvious,” and the one thing people get wrong costs them real money in the estimate. Let me decode the meter, then walk a worked bill.

The pricing model, decoded

Langfuse has four managed tiers, plus a free self-host under MIT.

Tier	Price	Included	Overage	Retention
Hobby	$0	50k units/mo (hard cap)	none	30 days
Core	$29/mo	100k units	$8/100k	90 days
Pro	$199/mo	100k units	$8/100k	3 years
Enterprise	$2,499/mo	100k units	$8/100k	custom

The quota unit is the “billable unit,” and this is the part everyone misreads. A billable unit is not one trace - it is traces plus observations plus scores, combined. A single trace of a multi-step agent can contain a dozen observations, one for each LLM call, tool call and retrieval step. Every score you attach adds units too.

So if you budget by trace count, you will under-estimate. The overage rate steps down from $8 to $6 per 100k at higher volume, which softens the top end, but the unit definition is what bites first.

The Hobby tier’s 50k units is a hard cap, not a soft overage - traffic past it stops being recorded rather than billing you. Pro at $199/mo is mostly about retention (3 years) and compliance reports (SOC2, ISO27001, a BAA), not raw capacity. If you do not need those, Core covers a lot of ground.

Estimate your bill

Say you run an agent that averages 8 observations per trace and you score every output once. That is 8 plus 1 plus 1, so roughly 10 billable units per user request.

• 100k requests a month is about 1M units. On Core that is 900k units of overage past the first 100k, at $8 per 100k, so roughly $29 + $72 = about $101/mo. That matches Langfuse’s own “$101/mo at 1M events” figure.
• 500k requests is about 5M units, or roughly $400 to $430/mo as the overage rate steps toward $6.
• Under about 5k requests a month you fit inside the free Hobby cap and pay $0.

The lever you control is observations per trace. A chatty agent with deep tool trees multiplies your unit count. A simple single-call chatbot barely moves it. Estimate your ratio first, then multiply.

The trap - “free self-host” means free software, not free ops

The MIT self-host is real and generous. Only three features are enterprise-gated, so tracing, evals, prompt management, human annotation and RBAC are all free when you run it yourself. At any serious volume, self-hosting beats every managed tier because the per-unit cost disappears.

But free software is not free operations. Langfuse v3 moved from “just Postgres” to Postgres plus ClickHouse plus Redis plus S3-compatible storage - four services to stand up and keep running. The migration to that architecture is where self-hosters get stuck, with GitHub threads reporting memory errors and timeouts. Budget the engineer-hours, not just the server bill. If you cannot spare the ops capacity, the $29 Core cloud tier sidesteps the whole thing.

One more fact to file away - Langfuse was acquired by ClickHouse in January 2026. The MIT license and self-host survived, but you are now betting a multi-year stack on a database company’s subsidiary. Full details are on the Langfuse page.

Cheaper picks, and when they win

Langfuse is already cheap, but two comparisons are worth knowing.

If you want managed cloud for even less, look at Opik. Comet’s platform bills spans, not billable units, and its Pro cloud tier is $19/mo for 100k spans - the cheapest paid cloud of the major eval platforms, with $5 per additional 100k. Its OSS build is Apache-2.0 with the full feature set self-hosted, unlimited, no gates. If you would rather not run four services and want the lowest managed price, Opik is the move. The watch-out is per-seat pricing that climbs with team size.

If you are comparing against LangSmith, Langfuse is the cheaper side by a wide margin. LangSmith bills base traces at $2.50 per 1,000 and runs roughly $2,514/mo at 1M base traces on one seat - about 25x Langfuse for comparable volume - and you cannot self-host below Enterprise to escape it. If price at scale is why you are reading a pricing guide, that gap is the headline.

Bottom line

Langfuse pricing is honest and cheap, and the only real gotcha is the billable-unit definition - budget for observations and scores, not just traces. Self-host free if you have the ops capacity, take Core at $29/mo if you do not, and know that either way you are paying a fraction of LangSmith at scale. If you want managed for even less, Opik at $19/mo undercuts it.

All figures here were read from each vendor’s own materials. This category ships breaking changes monthly, so we re-verify every 30 days.