The Langfuse Free Tier Explained in 2026 - Limits, Cap and When to Leave
Langfuse has two free options, and they are not the same. Here's what the Hobby cloud tier's 50k-unit cap really counts, how the free MIT self-host differs, and when Opik's free tier beats both.
Published:
“Is Langfuse free” has a more interesting answer than yes or no. There are two free Langfuse options, they work completely differently, and picking the wrong one for your situation costs you either money or a weekend of ops. Let me explain what each one actually gives you, what the cap really counts, and when to jump.
The two free tiers are not the same
This trips people up constantly, so it goes first.
Option one is the Hobby cloud tier. It is $0, hosted by Langfuse, and capped. You get 50k billable units a month as a hard limit, 30-day retention, and 2 users. No infrastructure to run, no ops. When you hit the cap, you stop.
Option two is the free self-host. The self-hosted build is MIT-licensed and free, with no unit cap. You run it on your own infrastructure. The per-trace cost disappears entirely - you pay only for the servers.
Same product, two very different deals. Full detail is on the Langfuse page.
| Free option | Cost | Volume cap | Retention | Users | Ops |
|---|---|---|---|---|---|
| Hobby (cloud) | $0 | 50k units/mo, hard | 30 days | 2 | None |
| Self-host | $0 | None | You configure | Unlimited | You run it |
What “50k billable units” actually counts
The number 50k sounds generous until you understand the unit. A billable unit is one trace, one observation, or one score - not one request. A single user request in a real app often produces a trace with several observations (the spans for each LLM call and tool step) and a couple of eval scores on top. That one interaction can consume five, ten or more units.
So the practical read is this. 50k units does not mean 50k requests - for a multi-step agent it can mean far fewer. A simple single-call app will stretch the cap a long way. A verbose agent that traces every tool call and scores every output will reach 50k faster than the round number implies. Count your units per request before you assume Hobby covers your traffic. I broke the full quota math down in the Langfuse pricing guide.
And remember the Hobby cap is hard. It is not an overage rate. You do not pay a little more past 50k - you are simply done until next month or until you upgrade.
When to leave the free cloud tier
Three signals tell you the Hobby tier has run its course.
You need more than 2 users. This is the most common one. The moment a second teammate needs their own login, the Core tier at $29/mo unlocks unlimited users and bumps you to 100k units with $8 per additional 100k.
You are hitting 50k units. If you are bumping the hard cap monthly, you either pay for Core or move to self-host. Which one depends on the next signal.
Your volume is high enough that even paid cloud stings. This is where self-host wins decisively. At 1M events a month, managed Langfuse runs about $101/mo, and self-host removes the per-trace cost entirely. If you have the ops capacity to run the self-host stack, it is the cheapest observability you can get - the software is free. The honest catch is that Langfuse v3 needs four services - Postgres, ClickHouse, Redis and S3-compatible storage - and the migration is where self-hosters get stuck. Free software, real ops time. See how to self-host Langfuse.
When Opik’s free tier beats Langfuse
Langfuse is not the only strong free tier, and for some teams Opik is the better free deal. It comes down to what you are short on.
On the cloud, the trade is volume versus seats. Langfuse Hobby gives you 50k units but only 2 users and 30-day retention. Opik’s Free Cloud gives 25k spans but up to 10 members and 60-day retention. So if you have a small team that needs several logins on a free plan, Opik’s 10 members beats Langfuse’s 2 outright - even though the raw volume is lower. If you are a solo builder pushing volume, Langfuse’s 50k wins.
On self-host, Opik has the cleaner license. Langfuse self-host is MIT with three features gated. Opik’s self-host is Apache-2.0 with the full feature set and no gates at all - unlimited spans, members and retention. If “open source with zero asterisks” matters to you, Opik is the more permissive choice. I compared them head to head in Opik vs Langfuse.
So which free tier should you start on?
- Solo builder, single app, want zero ops - Langfuse Hobby. The 50k units stretch furthest for one person.
- Small team that needs several free logins - Opik Free Cloud, for the 10 members and 60-day retention.
- You have ops capacity and want no caps - self-host Langfuse under MIT, or Opik under Apache-2.0 if you want zero feature gates.
- You have outgrown free volume but not ops - Langfuse Core at $29/mo for unlimited users, or Opik Pro at $19/mo for 100k spans.
The one mistake to avoid is treating the two Langfuse free tiers as interchangeable. Hobby is capped and hosted. Self-host is uncapped and yours to run. Know which problem you have before you pick. For the wider view, see the best free LLM observability tools and Langfuse alternatives.
Every price and limit here was read from each vendor’s own pages and verified on 26 July 2026. This category ships breaking changes monthly, so we re-verify every 30 days.
Frequently Asked Questions
Is Langfuse free?
Yes, in two different ways. The Hobby cloud tier is $0 with a hard cap of 50k billable units per month, 30-day retention and 2 users. Separately, the self-hosted build is free under the MIT license, with only three features enterprise-gated. Cloud free is capped and managed for you. Self-host free is uncapped but you run the infrastructure. They solve different problems.
What is a billable unit in Langfuse?
A billable unit is one trace, observation or score. So a single request that produces a trace with several spans and a couple of eval scores consumes more than one unit. The Hobby tier's 50k-unit monthly cap is a hard limit, not an overage tier, so a moderately busy app with multi-step traces can reach it faster than the round number suggests.
What happens when I hit the Langfuse free tier cap?
The 50k billable units on the Hobby tier is a hard limit, not a soft overage. You either upgrade to the $29/mo Core tier, which gives 100k units and unlimited users, or move to the free self-host, which has no unit cap at all. If you self-host, the per-trace cost disappears entirely - you only pay for infrastructure.
Is Langfuse or Opik better on the free tier?
It depends on team size and retention. Langfuse Hobby gives more monthly volume - 50k units versus Opik's 25k spans - but only 2 users and 30-day retention. Opik Free Cloud gives fewer spans but up to 10 members and 60-day retention. For self-host, Opik's Apache-2.0 build has no feature gates at all, while Langfuse gates three features. Pick by whether you need more volume or more seats.
Explore More
Tool Reviews
Related Articles
- Arize Pricing in 2026 - Phoenix Is Free, AX Is a Sales Call
- Braintrust Pricing Explained (2026) - The Processed-Data Trap
- Build vs Buy LLM Observability in 2026 - The Honest Decision Guide
- Galileo Pricing Explained (2026) - What You Actually Pay
- Helicone Pricing in 2026 - Decoded, and Why the Meter Is a Mystery
Free Newsletter
Get the LLM Evals Newsletter
Platform comparisons, pricing changes and eval technique deep-dives. No spam.
Related Articles
Evaluation of LLM Applications: A Practical 2026 Guide
A vendor-neutral guide to evaluation of LLM applications: metric selection, dataset sizing math, judge calibration, cost models and a tool comparison.
August 9, 2026
guideBLEU vs ROUGE vs BERTScore - Which to Use and Why All Three Fail on Chat
BLEU counts precision, ROUGE counts recall, BERTScore compares embeddings. Here is how each one actually computes a score, a worked example on the same sentence, and why none of them can grade an open-ended LLM answer.
July 28, 2026
guideContext Precision vs Recall Explained - Diagnosing RAG Retrieval in 2026
Context precision punishes noise, context recall punishes gaps. Here is how each retrieval metric is computed, a worked example, and how the two scores together tell you whether your retriever is over-fetching or missing documents.
July 28, 2026
Langfuse Review
Opik Review