# Observability instrumentation overhead

> In an AIMultiple benchmark updated 11 August 2026, instrumentation overhead measured near 0% for LangSmith, 5% for Laminar, 12% for AgentOps and 15% for Langfuse on a multi-agent travel-planning workload.

Measured runtime overhead added by LLM/agent observability tools.

- **Category:** Infrastructure
- **2026 frontier:** LangSmith near 0%, Laminar 5%, AgentOps 12%, Langfuse 15%
- **Credibility:** A single benchmark on one multi-agent travel-planning system, replaying 100 identical queries against an uninstrumented baseline. Overhead of this kind depends heavily on whether reporting is synchronous or asynchronous and on sampling rates, so treat the ordering as directional rather than as fixed constants.
- **Verification:** Single source
- **Source:** [AIMultiple agent monitoring benchmark](https://aimultiple.com/es/agentic-monitoring)
- **Last updated:** 2026-10-04