Send more, charge more is broken. We are changing it.

Metrics billable series is calculated based on two components: active series and data points per minute (DPM). By understanding these components, you’ll be able to better manage usage and reduce costs for your organization. See pricing docs –>

Yes, we provide volume discounts for enterprise-scale usage. Contact us for more details.

Process: The work done to receive, process and optimize the signals sent by your systems.

Write: The work done to write only the useful data to the database.

Retain: The work done to retain and protect your stored data over time.

Yes, we have additional retention options available. Contact us for more details.

Yes, we provide volume discounts for enterprise-scale usage. Contact us for more details.

Process: The work done to receive, process and optimize the signals sent by your systems.

Write: The work done to write only the useful data to the database.

Retain: The work done to retain and protect your stored data over time.

Yes, we have additional retention options available. Contact us for more details.

Yes, we provide volume discounts for enterprise-scale usage. Contact us for more details.

Process: The work done to receive, process and optimize the signals sent by your systems.

Write: The work done to write only the useful data to the database.

Retain: The work done to retain and protect your stored data over time.

Yes, we have additional retention options available. Contact us for more details.

Yes, we provide volume discounts for enterprise-scale usage. Contact us for more details.

A Host is any physical or virtual OS instance sending observability signals (including K8s metrics). It is considered active if it has sent a signal in the last 15 minutes.

Your bill is the sum of these active hours throughout the month—meaning you don’t pay for hosts that are spun down over the weekend or during off-peak hours.

Host hour billing applies when you opt into the curated experience built off the raw telemetry data.

Alternatively, you can opt out of the curated experience and be billed solely based on telemetry storage.

We made this change to give customers more control on telemetry usage and costs.

Customers can now use adaptive telemetry to reduce the unused / unwanted telemetry and lower their overall costs.

Read more in the documentation.

A Host is any physical or virtual OS instance sending observability signals (including K8s metrics). It is considered active if it has sent a signal in the last 15 minutes.

Your bill is the sum of these active hours throughout the month—meaning you don’t pay for hosts that are spun down over the weekend or during off-peak hours.

Host hour billing applies when you opt into the curated experience built off the raw telemetry data.

Alternatively, you can opt out of the curated experience and be billed solely based on telemetry storage.

We made this change to give customers more control on telemetry usage and costs. Customers can now use adaptive telemetry to reduce the unused / unwanted telemetry and lower their overall costs.

We made this change to give customers more control on telemetry usage and costs. Customers can now use adaptive telemetry to reduce the unused / unwanted telemetry and lower their overall costs.

A “session” is defined as the time that a user is spending in a frontend application on and off from start to finish or timeout. A user session is either created or resumed when a user visits and uses the site. A user session ends after a maximum lifetime of 4 hours or after inactivity timeout of 15 minutes. The SDK will automatically create a new user session in any of these situations, if the user returns.

Yes, we provide volume discounts for enterprise-scale usage. Contact us for more details.

A database host is any database instance sending monitoring data to Grafana Cloud. It is considered active if it has sent data in the last 15 minutes.

Your bill is the sum of active host hours throughout the month, meaning you don’t pay for database instances that are spun down over the weekend or during off-peak hours.

Each database server instance counts as one database host – for example, each MySQL or PostgreSQL server, each RDS or Cloud SQL instance, or each node in a database cluster. If you run multiple database instances on a single machine, each instance is counted separately.

Yes, we provide volume discounts for enterprise-scale usage. Contact us for more details.

Database Observability generates metrics and logs that are stored in Grafana Cloud. These are billed at standard Grafana Cloud rates for metrics and logs, separate from your host hour usage. You can control telemetry volume by adjusting your collection configuration in Grafana Alloy.

An active AI user is anyone who uses Grafana Assistant during the monthly billing period — sending a message, clicking an action like Explain in Assistant, or working with Assistant through any of its surfaces: in Grafana or Workspace, in Slack or Microsoft Teams, in the Grafana mobile or desktop app, through the gcx CLI, or via Assistant’s API. Connecting through the Cloud MCP server also counts as active use.

Each active user includes up to 40M tokens per month. Additional usage is charged at $2 per 1M tokens. Token limits apply per user and are not shared across users.

Token consumption falls into two buckets:

• Your per-user tokens (40M per user per month) cover everything you do with Assistant manually — conversations in any surface, Assistant Search, dashboard and query generation, investigations you start, and any prompts or questions you send within a system-initiated investigation — plus Assistant Automations, which run on a schedule but count toward their creator’s tokens.

• The system-initiated pool (25M per month on Free and Pro; custom pools on contracted plans) covers all other AI work that runs automatically — investigations triggered by alerts, incidents, or IRM webhooks, LLM-based evals and guards from Agent Observability, and service-account or API usage.

Watcher agents are free while in public preview.

Once a pool is used, additional tokens are billed at $2 per 1M.

Tokens consumed by system-initiated AI activity — investigations triggered automatically by alerts, incidents, watchers or IRM webhooks, LLM-based evals and guards, and service-account or API usage — draw from a shared monthly pool: 25M tokens on Free and Pro plans, with custom pool sizes on contracted plans, separate from per-user allowances. The pool is shared across all system-initiated AI usage in your stack. Assistant Automations are the exception: they run automatically but draw from their creator’s per-user tokens, not this pool. (This pool replaces the 25M tokens previously described as “service account usage.”)

Not for data access. When your own coding agent or MCP client uses gcx or the Cloud MCP server to query Grafana — pulling metrics, dashboards, alerts, or incidents — the reasoning happens in your agent’s model, and no Grafana AI tokens are consumed. Grafana tokens are consumed when Grafana’s models do the work: chatting with Assistant through the gcx CLI, invoking Assistant tools like ask_assistant through MCP, or starting an investigation from either surface. One thing to know: connecting through the Cloud MCP server counts you as an active AI user for that billing period, even if your token usage is zero.

Searching itself doesn’t. The retrieval — semantic and keyword search across your dashboards, incidents, investigations, alert rules, and infrastructure — runs on Grafana’s search index, not an AI model, and consumes no tokens. Tokens are consumed by the AI work around a search: a small amount to rewrite your query for better results, and — the bulk of it — when Assistant reasons over what comes back, either as part of its reply in chat or through Deep Search’s enrichment, which runs a full Assistant pass on each new search. There is no separate search-token pool; this usage draws from the same allowances as the rest of Assistant.

A token is a piece of text that an AI model reads or generates. Tokens can be whole words, parts of words, punctuation, or spaces. As a rough rule of thumb: 1 token ≈ 4 characters in English; 100 tokens ≈ 75 words. Tokens can be consumed by active AI users and by system-initiated usage; system-initiated usage is not billed per user.

Yes, we provide volume discounts for enterprise-scale usage (note: discounts do not apply to tokens). Contact us for more details. Contact us for more details.

An investigation is a longer, deeper task you hand to Grafana Assistant: point it at an alert, incident, or question, and it forms hypotheses, dispatches sub-agents to prove or disprove them against your telemetry, and produces a structured report with findings, supporting evidence, and recommended next steps.

Investigations are billed by the tokens they consume — there is no per-investigation or per-seat charge. Investigations you start yourself count toward your per-user Grafana Assistant allowance (40M tokens per user per month). Investigations triggered automatically — by alerts, incident updates, watchers or IRM webhooks — count toward the system-initiated token pool (25M tokens per month included). If you send prompts or questions within an automatically triggered investigation, those interactions count toward your per-user allowance. Beyond your included allowances, additional tokens are billed at $2 per 1M.

Tokens consumed by system-initiated AI activity — investigations triggered automatically by alerts, incidents, or IRM webhooks, LLM-based evals and guards, and service-account or API usage — draw from a shared monthly pool: 25M tokens on Free and Pro plans, with custom pool sizes on contracted plans, separate from per-user allowances. The pool is shared across all system-initiated AI usage in your stack. Assistant Automations are the exception: they run automatically but draw from their creator’s per-user tokens, not this pool. (This pool replaces the 25M tokens previously described as “service account usage.”)

It depends on the scope of the problem and the effort level you choose — deeper investigations explore more hypotheses and query more data.

A token is a piece of text that an AI model reads or generates. Tokens can be whole words, parts of words, punctuation, or spaces. As a rough rule of thumb: 1 token ≈ 4 characters in English; 100 tokens ≈ 75 words.

Yes, we provide volume discounts for enterprise-scale usage (note: discounts do not apply to tokens). Contact us for more details. Contact us for more details.

A generation is a single call to an AI model — one request and response — captured by Agent Observability. Agents typically produce multiple generations per user interaction: planning steps, tool selections, and final answers each count as one generation.

Token charges apply only to Evals & Guards that use an LLM to score or screen your application’s outputs (for example, LLM-as-judge evaluators). These tokens count toward your system-initiated token pool (25M tokens per month included); once the pool is used, additional tokens are billed at $2 per 1M. Evals and guards that run static checks — schema validation, regex, heuristics — consume no tokens and incur no token charge.

Tokens consumed by system-initiated AI activity — investigations triggered automatically by alerts, incidents, or IRM webhooks, LLM-based evals and guards, and service-account or API usage — draw from a shared monthly pool: 25M tokens on Free and Pro plans, with custom pool sizes on contracted plans, separate from per-user allowances. The pool is shared across all system-initiated AI usage in your stack. Assistant Automations are the exception: they run automatically but draw from their creator’s per-user tokens, not this pool. (This pool replaces the 25M tokens previously described as “service account usage.”)

A token is a piece of text that an AI model reads or generates. Tokens can be whole words, parts of words, punctuation, or spaces. As a rough rule of thumb: 1 token ≈ 4 characters in English; 100 tokens ≈ 75 words.

Traces, logs, and metrics generated by your AI applications are stored in Grafana Cloud and billed at standard Grafana Cloud rates, separate from generations. You can control telemetry volume through your instrumentation and collection configuration.

Yes, we provide volume discounts for enterprise-scale usage (note: discounts do not apply to tokens). Contact us for more details. Contact us for more details.

A test execution is a Synthetic test running in a probe location per minute of run time. Example: A synthetic test running in 5 locations that takes 1.5 minutes to complete will consume 10 executions each time it runs.

To estimate the number of test executions per month, you will need the following inputs:

• Number of Locations (Probes)
• Number of Targets (Tests)
• Duration of Test (rounded to the nearest minute)
• Frequency of Tests (in Minutes)

Yes, we provide volume discounts for enterprise-scale usage. Contact us for more details.

A test execution is a Synthetic test running in a probe location per minute of run time. Example: A synthetic test running in 5 locations that takes 1.5 minutes to complete will consume 10 executions each time it runs.

MultiHTTP checks are like HTTP checks but you can test multiple URLs in a single check to validate multi-step user flows. k6 scripted checks are like MultiHTTP checks, but you can write the checks in code instead of via a form-based approach.

To estimate the number of test executions per month, you will need the following inputs:

• Number of Locations (Probes)
• Number of Targets (Tests)
• Duration of Test (rounded to the nearest minute)
• Frequency of Tests (in Minutes)

Yes, we provide volume discounts for enterprise-scale usage. Contact us for more details.

VUh is calculated by using the following formula:

(Maximum number of VUs x test execution duration in minutes) / 60 minutes = VUh

If your test ramps up to a maximum of 100 VUs and runs for 10 minutes, the test run will use 16.67 VUh ((100 VUs x 10 minutes)/60 = 16.67 VUh).

Note that VU hours are calculated based on the period that the test run executes, not the preconfigured test duration. If the test duration is 2 or more hours but runs for only 30.01 minutes, then the subscription will be charged for the test-execution period rounded up to the next minute (in this case, 31 minutes).

The minimum for a test is 1 VUh. Find additional details here.

Both the free tier and free trial period are limited to 500 VUh / month.

With our public calculator, you can model your testing workload and estimate the monthly and yearly consumption of VUH.

Yes, we provide volume discounts for enterprise-scale usage. Contact us for more details.

Any user that logs into Grafana during the billing month is considered an active user.

Enterprise plugins are a paid add-on for the Cloud Pro Pay As You Go Plan that is priced at $55 per active user per month. By default, your account will have Grafana without Enterprise plugins, which is priced at $8 per active user per month.

Yes, we provide volume discounts for enterprise-scale usage. Contact us for more details.

Any user that logs into Grafana during the billing month is considered an active user.

Enterprise plugins are a paid add-on for the Cloud Pro Pay As You Go Plan that is priced at $55 per active user per month. By default, your account will have Grafana without Enterprise plugins, which is priced at $8 per active user per month.

Yes, we provide volume discounts for enterprise-scale usage. Contact us for more details.

An active IRM user is defined as a user who is included in OnCall schedules or escalation chains or performs any of the following actions:

• Changed the status of an alert group or OnCall configuration
• Received a page or paged another user
• Created, edited, or updated an incident

Yes, we provide volume discounts for enterprise-scale usage. Contact us for more details.