

LangSmith
LangSmith pricing charges $39 per seat a month, then $0.005 for every trace past your allowance, then another $0.0025 if you want that trace to outlive 14 days.
Updated on:
LangSmith pricing: you pay twice for the same trace
LangSmith pricing charges $39 per seat a month, then $0.005 for every trace past your allowance, then another $0.0025 if you want that trace to outlive 14 days. The second charge is the one buyers miss. LangChain the framework stays free and MIT-licensed. LangSmith, the platform watching what the framework does, is where the meter runs.
Key takeaways
A base trace costs 0.005 LSU and one LSU is $1.00, so a trace is $0.005, or $5.00 per 1,000.
Keeping a trace past 14 days costs an extra 0.0025 LSU, so a long-retention trace lands at $0.0075, or $7.50 per 1,000.
Base traces got ten times more expensive in nine months, from $0.0005 in December 2025 to $0.005 today, while the retention upgrade got cheaper.
Online evaluators and automation rules upgrade traces to extended retention by default, so the bill moves when you turn a rule on, not when you touch a pricing setting.
LangSmith pricing in 2026
Tier | Price | What's included | Metered limit
|
|---|---|---|---|
Developer | $0 per seat | 5,000 base traces a month, 1 seat, community support, cloud only | Hard-capped at 5,000 traces until you add a card |
Plus | $39 per seat a month | 10,000 base traces a month, unlimited seats, 1 free small serverless deployment, Engine and Fleet | 500,000 trace events an hour, 5.0 GB an hour |
Enterprise | Custom | Self-hosted, hybrid or BYOC, custom SSO, ABAC, RBAC, support SLA | Custom |
What LangSmith actually meters
One trace is one execution of your application, and LangChain defines it generously. Every LLM call, tool call and retrieval step inside a single agent invocation rolls up into one billable trace, capped at 25,000 runs before LangSmith rejects the rest. A 200-step agent loop costs the same $0.005 as a one-shot completion, which favours anyone running deep agents.
The second meter is time. Base traces live 14 days, extended traces 180, and the upgrade costs 0.0025 LSU on top of the base charge. Your invoice carries two lines, all traces and extended retention upgrades, because a trace recorded in June can be upgraded in July.
You don't choose every upgrade. An online evaluator that scores a trace upgrades it. An automation rule matching a single run upgrades the entire trace, and a thread-level rule upgrades every trace in the thread. Retention extension is on by default for both. UI feedback, notes and annotation queue adds leave the tier alone, and experiments start at extended retention.
Everything else rides other units. Deployments, Sandboxes, Engine and Fleet meter in LangChain Compute Units at $1.50 each, storage in LangChain Storage Units at $1.00. Engine burns roughly 5 to 30 LCUs per run, so $7.50 to $45, every six hours unless an admin disables it.
How credits work
Gateway Credits are a separate prepaid balance buying model calls, not traces.
Route model traffic through the LLM Gateway without your own provider key and LangChain bills the tokens at face value, charging the spend as LCUs against a balance you buy upfront. The purchase carries a processing fee LangChain never publishes a rate for, and the minimum top-up is $4.50, which is three LCUs. A gateway request still costs a trace, with no extra LangChain fee on top.
Expiry, rollover and pooling rules are not documented on the pricing page or in the billing docs. Enterprise commitments differ: LCU usage draws down prepaid LangChain credits on a Credit usage tab. VC-backed startups can apply for up to $10,000 in credits.
What happens when you hit the limit
It depends which limit, and LangSmith has three kinds.
Exceed your included traces and nothing stops. LangSmith auto-charges and bills in arrears. There's no spend cap, only usage limits you set yourself, which LangChain warns are approximate.
Exceed a self-set usage limit and you get a 429. The extended-retention limit cuts deeper: once reached, LangSmith blocks anything that would create another long-lived trace, so automation rules and evaluators with retention extension enabled stop running.
Exceed the plan rate limits and you get a 429 too. Developer accounts without a card cap at 50,000 trace events an hour and 5,000 traces a month. A card lifts that to 250,000. Plus gets 500,000.
How LangSmith pricing has changed across all these years
Date | Milestone | Source
|
|---|---|---|
1 Oct 2026 | Deployment customers on old per-run and uptime pricing move to LCU and LSU metering | Vendor |
14 Sep 2026 | Maximum extended retention on SaaS cut from 400 days to 180. Self-hosted and BYOC unchanged | Vendor |
24 Aug 2026 | UI feedback and reviewer notes stop upgrading traces to extended retention | Vendor |
17 Aug 2026 | Gateway Credits minimum purchase drops to $4.50, equal to three LCUs | Vendor |
31 Jul 2026 | Rates redenominate into LSUs. Base trace doubles to 0.005 LSU, upgrade holds at 0.0025 LSU | Vendor |
14 Jun 2026 | LCU appears on the pricing page at $1.50, trace rates still quoted in dollars | Vendor |
2 Feb 2026 | Both plans settle at $0.0025 per base trace and $0.0025 per upgrade | Vendor |
24 Jan 2026 | Plans split: Developer pays $0.005 per base trace, Plus $0.0025 | Vendor |
31 Dec 2025 | Base traces $0.50 per 1,000, extended upgrade $4.50 per 1,000 | Vendor |
Customer
Sentiment Highlights
"LangSmith charges you for seats and per trace. They charge you even more if you want to keep a trace for more than 14 days"
NeutralCrane, Hacker News, August 2025
"gets quite expensive quite fast"
stingraycharles, on the prompt engineering workflow, Hacker News, December 2025
Explore other providers

Intercom Fin
Chatbot
Intercom Fin pricing charges $0.99 per outcome, which makes it the clearest outcome-based model in this index and the one every other support vendor gets compared against.

Ideogram
Image / Video Generation
Ideogram pricing sells a monthly bucket of priority credits, and priority buys speed rather than access.

Vertex AI
Enterprise LLM
Vertex AI pricing sells the same Gemini model at four different per-token rates and a fifth rate that's measured in reserved capacity per hour.
How much does LangSmith cost per trace?
Is LangSmith free?
What counts as one LangSmith trace?
Does self-hosting LangSmith change the pricing?
























