LangSmith Trace Spend & Retention Checklist

Look forCount every distinct organization user, including invited users, at US$39 per seat per month on Plus. Seats bill on the 1st. LangSmith prorates seats added mid-month and gives no credit for seats removed.

Look forA trace is one execution of the application, however many runs it contains, up to 25,000. Plus includes 10,000 base traces a month. Each base trace past that costs 0.005 LSU, billed monthly in arrears. Developer includes 5,000 traces a month and stops there until someone adds a card.

Look forBase keeps traces 14 days. Extended keeps them 180 days for an extra 0.0025 LSU per trace. Set the tier per tracing project, or per environment in an agent-based workspace. A change applies only to new traces.

Look forCheck each online evaluator and automation rule for automatic retention upgrades. New rules and evaluators have the upgrade on by default. Thread-level rules upgrade every trace in the thread. API feedback upgrades only with extend_trace_retention. UI feedback and manual annotation-queue additions leave retention unchanged. Experiment runs start at extended retention.

Look forAn automation rule with a sampling rate or a filter, for example 10% of traces or every errored trace, upgrades only that slice. For anything needed past extended retention, a rule that copies inputs and outputs into a dataset keeps them indefinitely.

Look forBudget from LangChain's pricing page, which is binding. The usage and billing docs list no per-trace price and point to it. On October 12, 2026 the page lists 0.005 LSU per base trace past the included volume, which is US$5 per 1,000 at US$1 per LSU. Upgrading a trace costs 0.0025 LSU, or US$2.50 per 1,000 for extended retention. Accounts from 2025 and August 2026 cite US$0.50 per 1,000 base traces and 400-day extended retention, the list price before 2026. LangChain cut SaaS extended retention to its current length on September 14, 2026. Check the first invoice against the page.

Look forOpen Usage limits under Billing and Usage in Settings. Enter a monthly workspace spend limit. LangSmith sets matching base and extended trace limits. At the extended limit, evaluators and rules that extend retention stop, and UI feedback and annotation queues keep working. In 2025, one developer hit this limit with code-evaluator experiments and could not start new experiments.

Look forLangSmith Deployment bills by vCPU-hour and GiB-hour. Engine costs about 7 to 45 LSU per run on a 6-hour schedule. Fleet bills past 37 included LSU a month on Plus, Sandboxes past 8 LSU, and the Perceived Error Tuned Evaluator at 0.015 LSU per evaluation run. On Plus, LSUs bill on top of seats. On Enterprise, they draw down committed credits.