Salesforce Agentforce Pricing: Flex Credits, Agentforce Cost per Action and the Line Nobody Budgets

Salesforce Agentforce pricing has two live models and a third cost that neither of them measures. Flex Credits are 500 US dollars per 100,000 credits, a standard agent action costs 20 credits (about 10 cents) and a voice action 30 credits (about 15 cents). The older model charges 2 dollars per conversation. Seats run from 5 dollars per user per month up to Agentforce 1 editions from 550. All of those land on a Salesforce invoice. Money an agent causes to move outside Salesforce does not, and that is usually the larger number.

These rates were read from the Salesforce Agentforce pricing page on 16 September 2026 and are US list. Most partner write-ups get the headline right and then model the wrong unit, which is where budgets go wrong, so it is worth being careful about what each line actually charges for.

The actual Agentforce rates

Two billing models run alongside each other, and you should know which one your contract is on before you forecast anything.

What is the difference between Flex Credits and per-conversation pricing?

Per-conversation pricing charges a flat 2 dollars every time a conversation starts, whether the agent resolves the case or fumbles it. Flex Credits charge per action performed, at 20 credits each. The two models reward opposite behaviour. Per-conversation pricing is easy to forecast and pays the same for a resolution as for a dead end, so an agent that deflects nothing still costs full price. Flex Credits reward an agent that gets to the answer in three steps and quietly punish one that takes eleven.

The switch matters more than the rate. Per-conversation is bounded by how many people talk to your agent, which is a number you already know from your support volume. Flex Credits are bounded by how much work your agents do, which is the number you are actively trying to increase. That is a different shape of cost and it wants a different kind of control.

How many Flex Credits does an Agentforce action use?

Twenty for a standard action, thirty for an Agentforce Voice action. At 500 dollars per 100,000 credits each credit is half a cent, so a standard action is about 10 cents and a voice action about 15.

The trap is the word action. One user interaction is rarely one action. A service agent that looks up a record, checks an order, drafts a reply and updates a case has performed four, so the interaction cost 40 cents rather than 10. Every Flow, prompt template and Apex method you attach to an agent is another thing it can spend a credit on, and richer agents naturally take more steps. When you model this, count actions per interaction rather than interactions, and be honest that the number rises as the agent gets more capable.

What Agentforce actually costs at production volume

Ten cents is the number that gets a pilot approved. It is not the number that shows up later. Take one service agent handling 2,000 interactions a day at four actions each: 8,000 actions, 160,000 credits, 800 dollars a day at list, which is roughly 24,000 dollars a month from one agent. A second agent doing half that volume adds another 12,000. Add Agentforce Voice, where each action is 30 credits rather than 20, and the same traffic costs half as much again.

Run that properly and the seat lines usually turn out to be the small half of the bill. A hundred Agentforce User Licenses at 5 dollars is 500 a month, a rounding error next to the meter. Even the 125 dollar add-on across fifty users is 6,250, still under a single busy agent at the volumes above. The Agentforce 1 editions bundle 2.5 million Flex Credits annually, which sounds enormous until you divide it by 20 and get 125,000 actions for the year, or about 342 a day. That is one moderately busy agent.

None of this makes Agentforce expensive. Per unit it is competitive, and per-action pricing is arguably fairer than charging for seats that sit idle. The point is that the cost is variable, it scales with exactly the thing you are trying to grow, and a forecast is not a ceiling.

Does Digital Wallet cap Agentforce spending?

No, and this is the single most important thing to understand before signing. Digital Wallet is Salesforce consumption tool and it is genuinely good: it shows credit balances and burn rate, projects when capacity runs out, and emails threshold alerts, commonly configured at 50, 75 and 90 percent, with more alerting available through Flow.

Every verb there is a reporting verb. Digital Wallet observes consumption and tells a human. It does not hold an action, refuse one, or switch an agent off when a number is reached, and Salesforce practitioners have made the same point publicly, noting the absence of any kill switch that freezes agents when a threshold hits. It is a smoke alarm, not a sprinkler. Useful, and not a limit, because a limit refuses.

That distinction decides how you should contract. If the only thing standing between a looping agent and a large overage is somebody reading an email on a Friday evening, negotiate the ceiling into the contract rather than relying on the console. Ask for a hard cap on overage, a defined rate for credits beyond the pack, and a written definition of what counts as an action, because that definition is what you are actually buying.

The cost line that is not on the invoice

The credit meter is the cost people model, because it lands on a Salesforce bill and somebody has to approve it. The larger exposure usually never gets there.

Agentforce agents are wired into the rest of the business on purpose. An action can call a Flow, an Apex callout or a MuleSoft connector, and any of those can reach a payables system, a supplier portal, an ads platform, a SaaS billing API or a Commerce Cloud order. If an agent renews an annual contract, tops up an ad balance or releases a batch of supplier payments, the Salesforce side of that transaction is a few cents of actions. The actual money left through your bank, and no credit report will show it.

We went looking for where that control would live and found that it does not. Searching all 1,757 standard objects documented in the Winter 27 Salesforce Object Reference on 16 September 2026, 178 objects carry money vocabulary and 41 carry agent vocabulary, and the intersection is exactly zero. Since policy in Salesforce is stored as records on objects, a missing object means the value has nowhere to live. The full measurement, including the control pass that proves the method is not blind, is on our Agentforce spend controls page.

How to budget for Agentforce without guessing

Four things make the difference between a forecast that holds and one that does not.

Model actions, not conversations. Instrument a pilot agent and count actions per resolved interaction, then use that multiplier. Teams that assume one action per conversation underestimate by three to five times.

Price the voice premium separately. Thirty credits against twenty is a fifty percent uplift on every action, so voice deployments need their own line rather than a blended rate.

Negotiate the definition of an action. The contract term you care about most is what counts as one. Get it written down, with the overage rate and a cap.

Put an enforced ceiling where the alert is. Digital Wallet tells you. Something has to refuse. For consumption that means a contractual cap; for money the agent moves outside Salesforce it means a policy layer with real agent spend controls: a budget that aggregates across a month rather than resetting, a counterparty allow list, and an approval threshold that holds a payment for a named human above an amount you set.

It is also worth separating the AI line from the operational one before you compare vendors. A lot of what an Agentforce service agent is bought to absorb is ordinary customer experience operations work, the onboarding, billing and document workflows behind the support queue, and knowing which of those you are actually paying to automate keeps the business case honest.

Is Agentforce worth it at these rates?

For a company already deep in Salesforce, usually yes, because the grounding in your own data and metadata is the hard part and Salesforce has it. Ten cents an action against the loaded cost of a human doing the same lookup is not a close comparison, and per-action pricing is a more honest model than charging for seats nobody uses.

The risk is not the rate, it is the absence of a valve. You are buying a variable cost with a meter, an alert, and no refusal, attached to agents that can also move money the meter never sees. Budget for the meter, contract for the ceiling, and put enforcement in front of anything that leaves your bank. Our comparison of AI agent payment platforms covers who does that job and how they differ.