LangSmith Pricing in 2026: Seats, Traces, and the New LSU Billing Model Explained

If you are budgeting for LangSmith in late 2026, your bill comes down to three moving parts: a fixed per-seat cost, an included monthly allowance of base traces, and usage-based charges for everything beyond that allowance. The complication is that LangChain repriced its usage billing in July 2026, moving trace overages and storage into abstract units called LangChain Storage Units (LSUs) at $1.00 each. That change explains why third-party guides still disagree about what a trace costs — and why a stale guide can leave you underestimating your overage bill by half.

Understanding LangSmith's Core Pricing Model

LangSmith's pricing splits cleanly into fixed and variable components. The fixed component is the seat: the Developer plan is free for a single solo user, while the Plus plan costs $39 per seat per month with no cap on how many seats an organization can add. The variable component is usage — traces beyond your included allowance, extended data retention, and, on paid tiers, deployed agents.

Both self-serve plans include a monthly allowance of base traces: 5,000 per month on Developer and 10,000 per seat per month on Plus. Once you exceed that allowance, you pay as you go. The Plus plan also unlocks access to Deployment, Engine, and additional services, which introduce their own metered costs. Enterprise pricing is negotiated directly with LangChain and is outside the scope of this breakdown.

One boundary worth stating up front: LangSmith's charges cover observability, evaluation, and deployment infrastructure — not the LLM API calls themselves. Token costs from providers such as OpenAI or Anthropic are billed separately by those providers and sit on top of whatever LangSmith charges.

LangSmith Developer vs. Plus: Feature and Cost Comparison

The two self-serve plans are aimed at different stages of the same journey: Developer for solo experimentation, Plus for teams building and deploying agents. The table below juxtaposes the key quantitative and qualitative features side by side, consolidating the official plan structure with the current published figures.

Two practical consequences follow from this structure. First, because Plus trace allowances scale per seat, a ten-person team on Plus starts with 100,000 included base traces per month before any overages apply. Second, because Developer is capped at one seat, there is no path to add teammates without moving to Plus — the free tier is genuinely a single-user tier, not a per-user free allowance within a team account.

Demystifying Usage-Based Costs: Traces, LSUs, and LCUs

Usage-based charges are where budgets go wrong, because LangChain now bills usage through abstract units rather than a simple per-trace sticker price. The unit that matters most for traces and storage is the LangChain Storage Unit. As ZenML's pricing guide by Hamza Tahir explains, one LSU costs $1.00 and covers data stored as traces, in deployment databases, and in sandbox storage.

For traces specifically, the conversion is what determines your effective per-trace price. Traces beyond your plan allowance are billed at 0.005 LSU each, per pricing documentation captured by UsagePricing's Abhilash John in July 2026. At $1.00 per LSU, that works out to $5.00 per 1,000 overage traces — a figure you will not find printed as a per-trace price on the current pricing page, which is exactly why so many third-party guides disagree with one another.

Extended data retention adds a second variable. Retention beyond the standard 14-day window is available as an add-on on both plans, and traces kept under extended retention are billed at a higher rate than base traces — $5.00 per 1,000, versus the $2.50 per 1,000 that applied to base traces before the July 2026 reprice. In other words, how long you keep a trace can matter as much as how many traces you generate.

Trace Overage Costs: Reconciling Conflicting Information

If you have seen three different prices for LangSmith traces, all three were real at different moments. The reconciliation is chronological:

• $0.50 per 1,000 base traces — the legacy price that many older guides still cite. Inference.net's June 2026 analysis explicitly warns that guides quoting this figure are stale.
• $2.50 per 1,000 base traces — the price Inference.net verified against the official pricing page in June 2026, and the pre-July 2026 baseline that UsagePricing documents.
• $5.00 per 1,000 base traces — the effective current rate after the July 2026 reprice, derived from the 0.005 LSU per overage trace conversion at $1.00 per LSU.

The practical implication for budgeting is significant: a team that budgeted from a stale guide at $0.50 per 1,000 traces would underestimate current overage costs by a factor of ten. A team that budgeted from a mid-2026 guide at $2.50 per 1,000 would still be off by half. The number to plan around as of late 2026 is $5.00 per 1,000 base traces beyond your allowance, with extended-retention traces carrying the same effective rate as a distinct, higher-cost line.

It is worth noting that this LSU-based figure is derived from conversion rates documented by secondary sources referencing the official pricing page, rather than from a single printed per-trace price — so treat it as the current effective rate, subject to change by LangChain.

How LangChain Storage Units (LSUs) and Compute Units (LCUs) Impact Your Bill

Deployed agents introduce a second billing surface beyond traces. According to LangChain's own billing documentation, LangSmith Deployment charges based on the resources a deployment consumes: requested vCPU and memory per running replica, with autoscaling adding billable replicas. Dedicated deployments run — and bill — continuously, while serverless deployments stop charging compute at scale-down rather than when traffic stops, a distinction that matters for spiky workloads.

Database resources are metered separately, with dedicated database vCPU and memory billed in LSUs. Notably, database storage and the serverless shared database are not billed, and managed Redis is not billed on any deployment type or size — so the metered costs concentrate on compute and database compute, not raw storage.

One genuine ambiguity remains in the third-party record: ZenML's guide describes deployment compute being measured in LangChain Compute Units (LCUs), while LangChain's own documentation states that deployment compute is measured in LSUs. The official documentation is the more authoritative source here, but if you are modeling deployment costs, be aware that some analyses use the LCU terminology for the same underlying vCPU-and-memory metering.

Estimating Your LangSmith Bill: Key Takeaways for Budgeting

To accurately budget for LangSmith, evaluate your team's size for seat costs, estimate your monthly trace volume to account for overages, and consider needs for extended data retention and deployments which consume LSUs and LCUs.

Concretely, a five-person team on Plus pays $195 per month in seats and starts with 50,000 included base traces. If that team generates 150,000 base traces in a month, the 100,000 overage traces at $5.00 per 1,000 add roughly $500 — meaning usage, not seats, dominates the bill once volume grows. A solo developer on the free tier can run 5,000 traces a month at zero cost before the same per-trace metering kicks in.

Monitor your LangSmith dashboard for actual trace volume, LSU consumption, and LCU usage against your initial estimates to refine your budget. The measurable signal to watch is the ratio of billed overage traces to your included allowance: once monthly overages consistently exceed your included base-trace allowance, seat-level upgrades or retention-policy changes will move your bill far less than trace-volume management will.

Sources

• LangSmith price change — LangSmith reprices usage into LCU/LSU un — Usagepricing
• Manage billing in your account - Docs by LangChain
• LangSmith Plans and Pricing — Langchain
• LangSmith Pricing Guide: How Much Does It Cost? - ZenML Blog
• LangSmith Pricing Explained (2026): Traces, Seats & Alternatives — Inference.net