LangSmith alternative built on OpenTelemetry
Last updated on
LangSmith is LangChain's platform for tracing, evaluating, and deploying agents. telemetry.dev collects traces, logs, and metrics from AI applications over OpenTelemetry and works out token usage and cost from them. The two overlap on tracing. LangSmith also covers evaluation and deployment, which telemetry.dev does not.
We checked the LangSmith facts on this page against its pricing page and documentation on October 15, 2026. Plans change, so confirm them in the sources at the end before you decide.
At a glance
| telemetry.dev | LangSmith | |
|---|---|---|
| What it covers | Traces, logs, metrics, token usage, and cost | Tracing, evaluation, prompt tools, and agent deployment |
| How you pay | One monthly price per organization | A monthly price per seat, plus usage |
| Free plan | $0 for 30,000 units a month, 2 seats, and 1 project | $0 for 5,000 base traces a month and 1 seat |
| First paid plan | Pro: $25 a month for 500,000 units and 5 seats | Plus: $39 per seat a month, with 10,000 base traces |
| For a larger team | Team: $50 a month for 2,000,000 units and unlimited seats | More Plus seats, or Enterprise at a custom price |
| Usage above the plan | Pro and Team keep ingesting, with no automatic overage fee | Billed pay as you go |
| What counts as usage | Each span, log record, or metric data point | Each trace, however many steps it has |
| Retention | 14, 30, or 90 days, depending on the plan | 14 days for base traces, or 180 days for extended traces, which cost more |
| OpenTelemetry | OTLP/HTTP traces, logs, and metrics, as protobuf or JSON | OTLP traces at https://api.smith.langchain.com/otel |
| Evaluation | None | Online and offline evaluation, datasets, annotation queues, Prompt Hub, and Playground |
| Hosting | United States | US or EU cloud, with hybrid and self-hosted options on Enterprise |
What a team would pay
The two products count usage differently. LangSmith counts traces. telemetry.dev counts every span, log record, and metric data point, so a trace with 10 spans is 10 units in telemetry.dev and 1 trace in LangSmith.
Take a team of 5 developers that sends 40,000 traces a month, each with 10 spans:
- telemetry.dev: 400,000 units, which fits in Pro at $25 a month.
- LangSmith: 5 Plus seats cost $195 a month. That includes 10,000 base traces, and LangSmith bills the other 30,000 at its pay-as-you-go rate.
Your result depends on how many spans each trace has, and on any logs and metrics you send to telemetry.dev, which count as units too.
When telemetry.dev fits
- You want one price for the organization instead of a price for each seat.
- You want OpenTelemetry logs and metrics next to your traces, not traces alone.
- You want no automatic overage charges when usage grows past the plan.
When LangSmith fits
- You test prompts and agents with evaluations and datasets.
- You collect human feedback in annotation queues.
- You deploy agents on LangSmith Deployment.
- You need your data stored in the EU, or you need to host the platform yourself.
- You need to keep traces for more than 90 days.
Send LangChain traces to telemetry.dev
telemetry.dev reads LangChain and LangGraph spans from OpenTelemetry instrumentation such as OpenLLMetry. The LangChain and LangGraph guide shows the setup. You can also wrap your model clients with the Python SDK or the TypeScript SDK to trace each model call a chain makes.
To compare the two products on the same data, send your spans to both at once. An OpenTelemetry Collector can export one stream of spans to LangSmith and to telemetry.dev, and LangSmith's OpenTelemetry guide shows that setup.
Sources
- LangSmith pricing, read on October 15, 2026
- LangSmith billing, for trace tiers and retention
- Trace with OpenTelemetry, for the LangSmith OTLP endpoint and the Collector setup
- telemetry.dev pricing
- telemetry.dev sub-processors, for the hosting location