iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
Salesforce AI is not billed through one universal “AI request” meter. Agentforce can charge for actions or conversations, Einstein features count prompts through the LLM gateway, and some products use a hybrid license-plus-consumption or business-metric model. Agentforce activity is converted to Flex Credits through the applicable rate card; Einstein Requests are rounded into 2,000-token units and then weighted by the prompt type. Testing, previews, sandboxes, and production use may all be metered, while unmetered operation is available only when Salesforce’s documented user and execution conditions are satisfied.
The practical consequence is that a forecast must separate action volume, conversation volume, prompt tokens, model multipliers, test traffic, Data 360 consumption, and API quotas. A request can be unmetered for the AI call and still consume credits in an underlying service, and an API can return an error because of a rate limit or timeout even when the billing balance is healthy.
What does Salesforce AI billing measure?
Salesforce documents three broad pricing models. The first is consumption-based pricing, where usage is measured in native units such as actions or prompts. The second combines per-user licenses with consumption charges. The third ties price to a business metric defined in the commercial agreement. The meter therefore depends on the feature and the order form, not simply on whether a user clicked an AI button.
| Pricing approach | Native meter | What to verify |
|---|---|---|
| Consumption-based | Actions, prompts, conversations, minutes, or another stated unit | Unit definition, rate-card conversion, included allowance, and overage terms |
| Hybrid license plus consumption | User entitlement plus a usage meter | Which users are licensed and which activity remains chargeable |
| Business-metrics-based | A metric defined by the commercial agreement | The metric calculation, reporting period, and contract-specific price |
Agentic usage means an autonomous agent can reason, perform multi-step tasks, and act independently. Salesforce counts that usage by actions. Embedded prompt-based capabilities are counted by prompts. Those are different meters even when both use a large language model.
#1 Best Overall
How does Agentforce convert activity into Flex Credits?
Agentforce converts metered native units into Flex Credits through the Agentforce Rate Card. Flex Credits are fungible across Actions, Prompts, Translations, and Voice Actions, so one balance can represent several Agentforce consumption categories. The public 2026 pricing page lists Flex Credits at $500 per 100,000 credits, equivalent to $0.005 per credit, but an organization’s order form and current rate card control its actual entitlement.
| Agentforce item | Published usage or price | Illustrative arithmetic using the public 2026 figures |
|---|---|---|
| Standard Agentforce action | 20 Flex Credits | 20 × $0.005 = $0.10 per action |
| Agentforce Voice action | 30 Flex Credits | 30 × $0.005 = $0.15 per action |
| Conversation | $2 per conversation | Listed as a separate conversation price, not as a stated Flex Credit amount |
The action arithmetic is an illustration of the public list figures, not a promise that every contract uses the same commercial terms. A forecast should multiply the expected number of standard and voice actions by their applicable credit values, then reconcile that result with the purchased Flex Credit pool. Conversations should be modeled separately because the public pricing page presents them as a dollar-per-conversation metric.
What counts as an action?
An action is the native meter for agentic work: an agent deciding to perform an operation, such as invoking an approved capability or completing a step in a multi-step task. The exact action catalog and any special rates come from the applicable Agentforce Rate Card. Do not assume that a long conversation, a single user message, and one agent action are interchangeable units.
The Tool Desk
Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Where can administrators see consumption?
Digital Wallet provides consumption visibility. Salesforce also identifies the AI Agent Generative AI Usage Data Model as a source for auditing metered and unmetered activity. Use both when reconciling a forecast with actual records, especially when test and production traffic share an org.
Are Einstein prompts unlimited?
No. Einstein Requests are metered calls through Salesforce’s LLM gateway unless the specific usage qualifies for unmetered treatment. Prompt usage is rounded up in 2,000-token chunks, combining input and output tokens for the request. A request with 6,500 total input and output tokens is counted as four prompts, not 3.25 prompts.
| Prompt category | Multiplier on the October 24, 2025 rate card |
|---|---|
| BYOLLM Starter Prompts | 4 |
| Salesforce-enabled foundational Basic Prompts | 4 |
| Standard Prompts | 10 |
| Advanced Prompts | 38 |
For budgeting, first count the rounded prompt units, then apply the multiplier for the prompt category, and finally convert the resulting usage under the current commercial rate card. The October 24, 2025 card states that usage types and multipliers can be updated, so procurement calculations should use the rate card attached to the active contract rather than treating these values as permanent prices.
How does token rounding affect a forecast?
Rounding creates a step function. A request just over 2,000 tokens is charged as two thousand-token units, while a request just under the next boundary avoids an additional unit. Measure both input and output tokens, group requests by prompt type, and model the distribution of request sizes instead of multiplying an average word count by an assumed one-to-one prompt rate.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Which Salesforce AI activity is metered?
Metering depends on the lifecycle phase and the feature. Salesforce says Agentforce Builder design and development features do not consume credits, but an actual chat or voice preview is metered. Testing through Agentforce Builder, Agentforce Grid, Testing Center, and a sandbox is metered, and production interactions are metered under the applicable consumption model.
| Activity | Documented treatment |
|---|---|
| Agentforce Builder design and development | Does not consume credits |
| Chat or voice preview | Metered |
| Testing through Agentforce Builder | Metered |
| Agentforce Grid | Metered in all lifecycle phases |
| Testing Center | Metered |
| Sandbox activity | Metered |
| Production interactions | Metered under the applicable consumption model |
A sandbox is an isolated environment, not an automatic billing exemption. Include scripted tests, previews, Grid runs, and regression suites in the same usage forecast as production unless the relevant contract explicitly says otherwise.
When is Salesforce AI usage unmetered?
Unmetered status is conditional, not a blanket property of an Einstein or Agentforce feature. Global unmetering requires all documented conditions to be true: the user has the Unmetered User-Based AI permission, execution occurs as the current user, and the call is not made by an automated process identity. If any condition fails, the usage is metered.
Salesforce documents a maximum of 30 LLM calls per minute per user for this unmetered path and notes that the ceiling is not currently technically enforced. That statement describes the documented allowance; it is not a guarantee that unlimited traffic is acceptable or that downstream services are free.
Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Repair Windows errors before they cause bigger problemsFix Now →Data 360 and other underlying services can consume their own credits even when the AI call itself is unmetered. Audit the downstream operation separately rather than labeling an entire workflow “free.”
Rank #4
What API limits apply to Salesforce AI?
Billing meters and API quotas are separate controls. An organization can have available credits and still receive a throttling or timeout response when it exceeds an API limit.
| Interface or operation | Documented limit | Failure behavior or scope |
|---|---|---|
| Models API, LLM generation | Generally 2,000 requests per minute per org per REST endpoint | Limits can vary by model; an exceeded limit returns HTTP 429 |
| Models API, embeddings or feedback | 1,000 requests per minute per org | An exceeded limit returns HTTP 429 |
| Apex methods in sandbox orgs | 500 requests per hour | Separate sandbox quota |
| Apex methods in demo and trial orgs | 150 requests per hour | Separate demo/trial quota |
| Agent API request duration | 120-second timeout | A timeout returns HTTP 500 |
The Models API limits are per organization and endpoint, so distributing calls across users does not create an unlimited pool. Salesforce directs customers to their account executive when a Models API increase is needed. Treat HTTP 429 as a rate-limit response, not proof that a credit balance is exhausted.
What does the Agent API timeout mean?
An Agent API call has a documented 120-second timeout. HTTP 500 in this context can mean the request exceeded that duration, so check elapsed processing time and request complexity before diagnosing the event as a general service outage. Government Cloud customers use api.gov.salesforce.com as the base endpoint.
How should you forecast Salesforce AI costs?
Use a meter-by-meter forecast rather than one blended “AI calls” estimate.
Best Value
- Classify each feature. Record whether it uses Agentforce actions, conversations, Einstein prompts, voice actions, translations, minutes, or a business metric.
- Map the commercial model. Identify included user licenses, purchased Flex Credits, conversation allowances, and any consumption that remains outside those entitlements.
- Apply the current rate card. For Agentforce, convert actions and other native units into Flex Credits. For Einstein, round total tokens to 2,000-token prompt units and apply the prompt multiplier. Use the active order form and rate card because multipliers and usage types can change.
- Separate environments. Estimate production, sandbox, Builder previews, Testing Center, and Grid activity independently. Grid usage is metered in every lifecycle phase.
- Include downstream services. Check Data 360 and other underlying-service credits even where an AI call qualifies as unmetered.
- Size the integration. Compare peak traffic with the per-org, per-endpoint Models API quotas, the sandbox or demo Apex quota, and the 120-second Agent API timeout. Build handling for HTTP 429 and timeout-related HTTP 500 responses.
How can teams audit usage and troubleshoot surprises?
Start with Digital Wallet to compare current consumption with the purchased Flex Credit pool. Then use the AI Agent Generative AI Usage Data Model to attribute activity to users, agents, environments, and usage states where the available records support that level of detail.
- If credits rise during a release rehearsal, inspect previews, sandbox tests, Testing Center runs, and Grid activity before blaming production traffic.
- If Einstein usage is higher than request counts suggest, examine token totals and the 2,000-token rounding boundary, then classify each request by its prompt multiplier.
- If users report failures while consumption looks normal, inspect API responses separately: HTTP 429 indicates a rate-limit condition, while an Agent API HTTP 500 can indicate the 120-second timeout.
- If a workflow is marked unmetered, verify the permission, current-user execution, and non-automated identity conditions for the specific call. A single missing condition makes that usage metered.
What should procurement compare before signing?
Two plans with similar user counts can produce very different bills if their meters differ. Compare the following items side by side:
- Action pricing versus conversation pricing.
- Included user licenses versus consumption entitlements.
- Flex Credits versus Einstein Requests.
- Prompt token volume and the multiplier for each prompt category.
- Data 360 or other underlying-service charges.
- Production, sandbox, preview, Testing Center, and Grid metering.
- Per-org and per-endpoint API limits, plus any documented increase process.
- Timeout behavior and the operational work needed to handle 429 and 500 responses.
Ask Salesforce to identify the exact rate card, effective date, included quantities, and treatment of test traffic in the order documents. Public list prices are useful for a first model, but they do not replace contract-specific entitlements and multipliers.
Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Scan for outdated or missing drivers - takes under a minute3Clear out junk files and repair common Windows errorsFrequently Asked Questions
Are Agentforce conversations converted into Flex Credits?
The public pricing page presents conversations at $2 per conversation and presents Flex Credits separately. Do not assume a conversation can be exchanged for a fixed number of Flex Credits; confirm the relationship, if any, in the commercial terms for your org.
How do prompt multipliers change an Einstein budget?
First round each request’s combined input and output tokens into 2,000-token prompt units. Multiply those units by the category multiplier on the rate card, then apply the contract’s conversion or allowance; a raw request count alone will miss both token rounding and category weighting.
Can an unmetered AI call still create a billable event?
Yes. Unmetered status applies to the qualifying AI call, not automatically to every service invoked by the workflow. Data 360 or another underlying service can consume its own credits and must be reviewed separately.
Is every Agent API HTTP 500 an outage?
No. Salesforce documents a 120-second Agent API timeout that returns HTTP 500. Check whether the request exceeded that processing window before treating the response as a platform-wide failure.
Free tools Windows power users keep installed
One-click scans. No signup required.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
