Langfuse is cheaper for almost every team, and the gap grows with headcount. LangSmith charges $39 per seat per month plus usage. Langfuse charges $29 per month flat with unlimited users, plus usage. The bigger difference is in what you’re paying for over time: how long your history lives, and whether you can take it with you.

LangSmith vs Langfuse pricing at a glance

LangSmith Langfuse
Free plan Developer: 1 seat, 5k base traces Hobby: 2 users, 50k units, 30 days
Entry paid plan Plus: $39 per seat per month, 10k base traces Core: $29 per month, 100k units, unlimited users, 90 days
Higher plans Enterprise: custom Pro $199 (3 years of data), Pro + Teams $499, Enterprise $2,499
Billed on Seats, plus traces and other usage Units: traces, observations and scores
Overage Pay-as-you-go in LangChain Standard Units (1 LSU = $1) $8 per 100k units, falling to $6 above 50M
Retention 14 days base; extended up to 180 days, at extra cost 30 days, 90 days or 3 years by plan
Bulk export Enterprise-only for sign-ups after 3 Aug 2026 Public API on every plan; scheduled blob export from Pro + Teams
Self-hosting Enterprise only Free, MIT core
Startup credits Up to $10k Not listed

Sources: LangChain pricing, LangSmith usage and billing, LangSmith data export, LangSmith changelog, Langfuse pricing, Langfuse self-hosting. Checked 9 Nov 2026.

How LangSmith pricing works

You pay per seat, then for usage. Plus includes 10k base traces a month across the plan. Beyond that, usage is pay-as-you-go. Since October 2026 it’s billed in LangChain Standard Units, at $1 per LSU, and the pricing page no longer prints a per-trace dollar rate. Check the rate on your own billing page before you forecast.

Retention is the line item to watch. A base trace is kept for 14 days. Anything you want longer is an extended trace, which costs more, and from 14 Sep 2026 the longest retention a SaaS customer can buy is 180 days, down from 400. Online evaluators, automation rules and some feedback settings can upgrade a trace to extended retention, so turning on evals can change the bill.

And if you signed up after 3 Aug 2026, bulk export is Enterprise-only. Teams that signed up earlier keep it on Plus until 1 Feb 2027.

How Langfuse pricing works

You pay a flat monthly fee for a bundle of units, then $8 per 100k units over it, falling to $7, $6.50 and $6 as volume grows. There’s no seat fee from Core upwards.

A unit is any tracing data point: a trace, a span, an event, a generation or a score. Each counts once, whatever its size. That makes the forecast simple, but you need your span count. An agent that makes nine LLM and tool calls per turn and scores each turn twice uses 12 units per turn, not one.

Retention comes with the plan rather than as a separate charge: 30 days on Hobby, 90 on Core, 3 years on Pro. And you can always self-host the MIT core for free, as long as you’re happy to run Postgres, ClickHouse, Redis and blob storage.

A worked bill

Say three people work on an agent that handles 20k turns a month. Each turn is one trace with nine spans and two scores, so 12 Langfuse units.

Double the team and Langfuse doesn’t move. LangSmith adds $117.

Which is better value?

The full comparison, beyond price, is in LangSmith vs Langfuse.

What neither price includes

Both bills buy you a recorder. Neither tells you whether your agent is any good: which traces your experts should label, whether your judge agrees with them, or which scores move the outcome your business runs on. That’s what TwoTail does, on top of either tool. It connects to Langfuse with a pasted key and no code change, and bills on the same idea Langfuse does: traces and observations. Basic starts at $69 a month for 250k units with unlimited seats.

Disclosure: I build TwoTail. Other comparisons: Braintrust vs Langfuse, Arize Phoenix vs Langfuse.