PE Collective
LangChain / LangSmith Pricing: What the Platform Costs
LangChain the framework is free and open source. LangSmith, the observability and evaluation platform, is where the pricing kicks in. If you're building production LLM apps with LangChain, you'll eventually want LangSmith for debugging and monitoring. Here's what that costs.
Developer (Free)
- ✓ 5,000 traces per month
- ✓ 1 user
- ✓ 14-day data retention
- ✓ Basic tracing and debugging
- ✓ LangGraph Cloud playground
Plus
- ✓ 100,000 traces per month included
- ✓ Additional traces at $0.50 per 1,000
- ✓ Up to 10 users
- ✓ 400-day data retention
- ✓ Evaluation datasets and experiments
- ✓ Prompt versioning and management
Enterprise
- ✓ Unlimited traces
- ✓ Unlimited users
- ✓ Custom data retention
- ✓ SSO and RBAC
- ✓ Self-hosted deployment option
- ✓ Dedicated support and SLA
- ✓ SOC2 compliance
LangSmith Pricing Tiers Explained
LangSmith's Developer tier is free and includes 5,000 traces per month, 1 user, and 14-day data retention. It is enough for local development and testing but not for production monitoring. The LangGraph Cloud playground is included, letting you test agent workflows visually before deploying.
The Plus tier at $39/user/month is where most teams land. You get 100,000 traces per month included, up to 10 users, 400-day data retention, evaluation datasets, and prompt versioning. The overage rate of $0.50 per 1,000 traces means a team running 500K traces/month pays $39 + $200 = $239/user/month. Watch this number closely once you hit production traffic.
Enterprise pricing is custom and includes unlimited traces, unlimited users, SSO, RBAC, SOC2 compliance, self-hosted deployment options, and dedicated support with SLA guarantees. If your trace volume exceeds 1 million per month across a team, Enterprise pricing typically becomes cheaper than Plus overage charges.
LangGraph Cloud: What Agent Hosting Costs
LangGraph Cloud is LangChain's managed hosting platform for LangGraph agents. It is separate from LangSmith and has its own pricing. The Developer tier is free and gives you one assistant with limited concurrent runs, enough for testing agent workflows.
The Plus tier at $35/month supports multiple assistants, higher concurrency, and longer-running agent executions. For production applications where agents run complex multi-step workflows (retrieval, tool calls, human-in-the-loop approvals), the $35/month covers moderate traffic.
At scale, LangGraph Cloud compute charges are the hidden multiplier in your LangChain bill. Each agent run consumes compute based on execution time and the number of steps. A simple two-step retrieval agent is cheap. A complex agent that loops through 15 tool calls, waits for human approval, and branches based on results can cost 10-20x more per run. The LLM API costs (from OpenAI, Anthropic, etc.) are on top of the LangGraph Cloud compute charges.
For teams evaluating total cost: a typical production setup with LangSmith Plus ($39/user/month), LangGraph Cloud Plus ($35/month), and moderate LLM API usage ($100-300/month) runs $175-375/month total. Compare this to building the same agent infrastructure yourself on AWS Lambda or Cloud Run, factoring in engineering time for deployment, scaling, monitoring, and error handling.
Hidden Costs & Gotchas
- ⚠ Overage charges on Plus hit fast. If you're running 500K traces/month, you'll pay $39 base plus $200 in overage ($0.50/1K traces). That's $239/user/month.
- ⚠ The framework itself is free, but LangChain apps tend to use more LLM calls than simpler architectures. Chain-of-thought, retrieval, and agent loops multiply your API costs.
- ⚠ Data retention on the free tier is only 14 days. If you need to debug an issue from last month, that data is gone.
- ⚠ LangGraph Cloud charges separately for agent compute. The Platform plan starts at $35/month for cloud-hosted LangGraph agents. Production deployments with high concurrency can run $200-500/month in LangGraph Cloud compute on top of your LangSmith subscription and LLM API costs.
- ⚠ LangGraph Platform pricing includes a Developer tier (free, single assistant, limited runs) and a Plus tier ($35/month) with more assistants and higher concurrency. Enterprise pricing is custom and includes dedicated infrastructure and SLA guarantees.
Which Plan Do You Need?
Solo developer prototyping
Free tier with 5K traces is enough for development and testing. You won't need more until you're handling real production traffic.
Small team shipping LLM features
Plus at $39/user/mo with 100K traces works well for teams of 2-5. Watch overage charges once you're past the initial 100K. If you don't need the observability, you can skip LangSmith entirely and just use the open-source framework.
Enterprise with multiple LLM apps
Enterprise is worth it once you have 10+ developers or need SSO and compliance. The unlimited traces alone can save money compared to Plus overage charges at high volumes.
The Bottom Line
LangChain is free. LangSmith Plus at $39/user/mo is worth it if you're debugging production LLM apps and need trace visibility. But watch the overage charges on traces. If you're building simple LLM integrations, you might not need LangSmith at all. Alternatives like Langfuse (open source) give you similar observability for free if you self-host.
Related Resources
Frequently Asked Questions
Is LangChain free?
LangChain the framework is completely free and open source. LangSmith, the observability and evaluation platform, has a free tier (5K traces/month) and paid plans starting at $39/user/month. You can use LangChain without LangSmith.
Do I need LangSmith to use LangChain?
No. LangChain works fine without LangSmith. LangSmith adds tracing, debugging, and evaluation tools that help in production. For prototyping and simple apps, the framework alone is sufficient.
How does LangSmith pricing compare to Langfuse?
Langfuse is open source and free if you self-host. Their cloud version has a free tier and paid plans that are generally cheaper than LangSmith. LangSmith has tighter integration with LangChain, but Langfuse works with any LLM framework.
What counts as a trace in LangSmith?
A trace represents one complete execution of your LangChain pipeline, including all the steps (LLM calls, retrieval, tool use). A single user request that triggers 5 LLM calls still counts as 1 trace. Complex agent loops with many steps are still 1 trace per top-level invocation.
Is LangChain worth using in production?
It depends on your use case. LangChain is great for complex chains, agents, and retrieval pipelines where you need the orchestration abstractions. For simple API calls to OpenAI or Anthropic, it adds unnecessary complexity. Many production teams start with LangChain and either stick with it or simplify to direct API calls.