---
title: "Agent Observability pricing | Grafana Cloud documentation"
description: "Learn how Grafana Agent Observability bills for generations, evaluation and guard tokens, and telemetry."
---

> For a curated documentation index, see [llms.txt](/llms.txt). For the complete documentation index, see [llms-full.txt](/llms-full.txt).

# Agent Observability pricing

Grafana Agent Observability bills on usage across three separate dimensions: the generations your agents produce, the tokens consumed by LLM-based evaluations and guards, and the standard telemetry your SDKs emit.

For current rates and included usage, refer to the [pricing page](/pricing/).

## How Agent Observability is billed

Three meters run independently:

- **Generations** for the AI model calls your monitored agents and applications make.
- **Evaluation and guard tokens** for evaluators that call an LLM.
- **Telemetry** for the OpenTelemetry metrics and traces the SDKs emit, at standard Grafana Cloud rates.

Generation usage is billed separately from token usage. OpenTelemetry metrics and traces aren’t included in generation pricing.

## Definitions

**Generation:** One call to an AI model made by your monitored agent or application. Each time your application calls an LLM provider and the SDK exports that interaction, it’s counted as a generation. One agent task can produce several generations if it makes several model calls.

**LLM-based evaluator:** An evaluation or guard that calls an LLM to judge or guard an interaction, such as an LLM judge evaluation.

**Deterministic evaluator:** An evaluation or guard that doesn’t call an LLM, such as regex, JSON schema, heuristic, redact, and tool filter rules.

**System-initiated token pool:** The stack-level token allowance that LLM-based evaluation and guard usage draws from.

## How usage is calculated

Generation billing covers ingesting and storing generation telemetry. It doesn’t cover the cost of running your model, which your LLM provider bills separately.

LLM judge evaluations and LLM-based guards consume tokens because Grafana runs an LLM to perform them, and that usage counts toward the system-initiated token pool for your stack. Deterministic evaluators and guards don’t consume tokens unless they call an LLM-backed evaluator.

Telemetry follows standard Grafana Cloud pricing and limits for [metrics](/docs/grafana-cloud/platform/pricing-and-usage/metrics/) and [traces](/docs/grafana-cloud/platform/pricing-and-usage/traces/). Agent Observability doesn’t add a separate usage cap for it.

## Special considerations

### When billing starts

Pricing and metering for generations and for evaluation and guard token usage start on October 1, 2026.

Telemetry billing follows standard Grafana Cloud pricing and is already active.

## View your usage

Agent Observability analytics views show generation volume, latency, errors, token usage, and cost patterns, which helps you identify expensive workflows.

For usage alongside the rest of your Grafana Cloud spend, go to **Cost Management and Billing > Usage** in your stack.

## Related

- [Pricing page](/pricing/)
- [Agent Observability documentation](/docs/grafana-cloud/observe-and-act/agent-observability/)
- [Optimize cost and performance](/docs/grafana-cloud/observe-and-act/agent-observability/guides/cost-optimization/)