A subscription with seats and AI calls included, plus US$3.00 per 1,000 stored AI traces, billed daily. Model inference runs on your own provider connections and each provider bills it to you directly, with no markup from us. Non-AI traces, and traces that were not stored, are not billed and are not credited.
10 seats included, no per-seat pricing
What you pay
Your providers
Your models
BYOK, at your own provider prices
+
Open Token plan
The platform
BYOK gateway, agent observability, spend control and finance. All included. Plus US$3.00 per 1,000 traces. Only stored AI traces count.
+
Optional add-on
Fraud detection
Add to any plan
Team
For a product team putting agents into production.
US$149/month
Annual (10% off): US$1,609.20/year, billed annually · save US$178.80
seats included
10seats included
AI calls / month
300kAI calls / month
per 1,000 stored AI traces
US$3.00per 1,000 stored AI traces
trace retention
14 dtrace retention
AI calls above the plan: US$0.12 per 1,000 calls
Extra seats: packs of 5 at US$35/month
Includes
BYOK gateway to 9 model providers
Agent monitoring: traces with cost, dashboards per agent
Budgets, alerts and loop detection
Finance: costs reconciled with your AWS invoice (CUR)
Hard limits and per-run stops in the request pathSoon
Items marked Soon are on our public roadmap and not yet generally available.
Every organization starts on Developer: free and basic
Up to 3 members, 1 project, 5,000 stored AI traces a month kept 3 days (metadata only) and 20 spans per second. Non-AI traces, and traces that were not stored, do not count. A card is required to activate the organization; nothing is charged until you choose Team or Business.
Add-ons
Add to any plan
For AI agents that move money. Add it to Team, Business or Enterprise.
Move the numbers to match your usage. The estimate uses the same rates as the plans above; your model providers bill inference to you separately.
Each request to a model through Open Token, the SDK or OpenTelemetry. Streams include 60 s; each extra 60 s counts as one more call.
One stored AI trace: a root run of your agent or app (a LangSmith root run or an OpenTelemetry root span) whose root was stored, however many steps it has. Each counts once. Non-AI traces, and traces dropped before storage, do not count and are not credited.
Estimate before taxes. Stored AI traces, and usage above your plan, are billed daily. Traces that are not AI, or that were not stored, are not billed and are not credited. Model inference is billed to you by your providers and isn't included.
What's included
Feature
Team
Business
Enterprise
BYOK gateway to 9 model providers
Agent traces with cost
14 days, up to 400
14 days, up to 400
Custom
Dashboards per project and agent
Budgets, alerts and loop detection
Finance: CUR reconciliation
FOCUS exports and chargeback
SSO
Data residency
US East
US East
Your AWS account or region
Fraud detection
Add-on
Add-on
Add-on
How it compares
A team of 7 with 400,000 calls and 80,000 traces a month pays about US$401 on Team. The same workload costs around US$198 with Langfuse plus a separate gateway, and about US$623 on LangSmith Plus (7 seats and traces at US$5.00 per 1,000 after the first 10,000), based on public prices checked in September and October 2026.
No. Models run on your own provider connections and each provider bills them to you at your own prices. Open Token charges a plan, US$3.00 per 1,000 traces, and AI calls and extended retention above the plan. We never sell or mark up model usage.
Is agent monitoring extra?
The features aren't: traces with cost, dashboards per agent, budgets and finance views are part of every plan. Stored AI traces are billed at US$3.00 per 1,000 on every plan, with 14 days of retention included. Non-AI traces, and traces that were not stored, do not count, and they do not reduce the bill.
How is the fraud add-on billed?
Monthly, on top of any plan: Growth US$1,500, Scale US$4,500 or Enterprise from US$10,000, or a 60-day pilot for US$3,000. The cost of its AI investigations is included. Overage is optional, with a monthly cap you set.
Is there a free plan?
Yes. Every organization starts on Developer, which is free, with up to 5,000 stored AI traces a month. Non-AI traces, and traces that were not stored, do not count. A card is required to activate the organization and to prevent abuse; nothing is charged on Developer. You're only charged if you upgrade to Team or Business: the plan, plus any usage above what it includes, under the plan terms. Team and Business also start with a 14-day trial with capped usage (30,000 traces on Team) and no usage charges.
How is usage charged?
Every day, for the previous day's stored AI traces and any AI calls above your plan, once the amount reaches the daily minimum. Smaller amounts roll over to the next day. Traces that are not AI, or that were dropped before storage, are not billed.
What happens if a payment fails?
We email the account owners and retry. If the invoice stays unpaid, the account moves to a restricted state and, eventually, new calls are paused until it's settled. Paying the open amount restores access right away.
Do you charge per seat?
No. Team includes 10 seats, Business 25 and Enterprise 50 or more. If you need more, add packs of 5 seats (US$35/month on Team, US$30/month on Business).
What counts as an AI call and a trace?
An AI call is a model request that goes through Open Token, where it's metered, attributed and checked against your budgets. Each call includes 60 seconds of streaming; every further 60 seconds counts as one more call. A billed trace is one stored AI trace: a root trace (a LangSmith root run or an OpenTelemetry root span) that includes an AI span and whose root was stored, with all its steps, however many spans it has. It counts once, in the month it is first stored, the same unit LangSmith bills. Plain HTTP traces, and traces held, sampled out, over a cap, rejected, or dropped before the root is stored, do not count. They do not reduce the bill.
Who can see which teams?
Owners and Finance always see every team. Admins, Developers and Viewers can be limited to specific teams, and to environments, when you invite them and later on Members. Once limited, they see only those teams' projects, and a limited admin manages only those teams' projects and their Developers and Viewers, not company-wide settings or billing.
How do invitations work?
The invitation email has a one-time link to set a password. The link works for 7 days and does not include a password. The person who sends the invite also sees a temporary password once on screen, to pass on if the email does not arrive. A pending invitation's name, email and role can be corrected, and the invitation sent again, without adding the person a second time.
See your AI spend the way finance does
Accounts are set up by our team. Tell us which model providers you use and we'll walk you through a live workspace with your own numbers.