Locus product guide · Updated November 10, 2026

AI agent pricing models: 6 ways to charge, with examples

AI agents are priced in six main ways: per seat, per token or API call, per action, per conversation, per outcome, or with prepaid credits, and many vendors mix two of them. Published 2026 prices include Salesforce Agentforce at $2 per conversation or $0.10 per action in Flex Credits, Intercom at $0.99 per Fin outcome, and Claude Enterprise at $20 a seat plus usage at API rates. Choose the unit your buyer can count and that moves with your cost.

What are the main AI agent pricing models?

Each model puts the cost of a long or failed agent run on a different party. Examples come from vendor pricing pages. Prices checked November 10, 2026.

What are the main AI agent pricing models?
Model	Billed unit	Published example	Who carries cost risk
Per seat	User per month	Agentforce Coworker: $50 a user a month, searches included	The vendor
Seat plus usage	User plus metered usage	Claude Enterprise: $20 a seat plus usage at API rates	Split between both
Per token or call	Model tokens or API calls	OpenAI GPT-6.1 Sol: $2 input and $10 output per million tokens	The customer
Per action	Each step the agent takes	Agentforce Flex Credits: 20 credits, or $0.10, per action	Mostly the customer
Per conversation	Each conversation	Agentforce: $2 per conversation	The vendor, within each conversation
Per outcome	Each finished result	Intercom Fin: $0.99 per outcome	The vendor, including failed attempts
Prepaid credits	Credits drawn per action	GitHub Copilot Pro: $10 a month with $15 of AI credits	The customer, up to the balance

Why are AI agents harder to price than other software?

An agent decides how many steps a task takes. The same request can take 3 tool calls or 30, so your cost per task swings while the customer sees one task.

Salesforce's own examples show the spread. A where-is-my-order request uses 2 actions, or $0.20. An appointment-scheduling flow uses 5 actions, or $0.50. A seat or per-conversation price hides that spread from the buyer, so the vendor absorbs it. A per-action or per-token price passes it to the buyer, who then needs spend limits.

How do the models compare on the same workload?

Take one customer whose agent runs 400 tasks a month for five people. Each task takes 3 actions and costs you $0.15 in model and API fees, and 60% end in a result the customer counts. Revenue below applies the published rates from the table above. Your cost is $60 in every case.

The rates come from different products, so the revenue gap reflects each vendor’s packaging more than the model itself. The useful comparison is risk. If this customer doubled its task count, seat revenue would stay at $250 while your cost rose to $120. Revenue per action, conversation, outcome and cost-plus would double with the cost.

How do the models compare on the same workload?
Model	Rate applied	Monthly revenue	Margin after $60 cost
Per seat	$50 × 5 users	$250	$190 (76.0%)
Per action	$0.10 × 1,200 actions	$120	$60 (50.0%)
Per conversation	$2 × 400 tasks	$800	$740 (92.5%)
Per outcome	$0.99 × 240 results	$237.60	$177.60 (74.7%)
Cost plus 30%	$0.15 × 1.3 × 400 tasks	$78	$18 (23.1%)

Which pricing model fits which kind of agent?

Match the model to who uses the agent and how much its runs vary.

An assistant employees use all day: a seat price with included credits, plus paid usage above the allowance.
A customer-facing support agent: per conversation or per outcome, with a monthly minimum.
Research, enrichment or data agents that call paid APIs: per action or credits, priced at cost plus a markup.
A platform that runs its customers' agents: pass token and API costs through with a markup, drawn from a prepaid balance.
Agents with long, unpredictable runs: any model, plus a hard cost cap per task.

How do I price agent tool calls per customer with Locus Pro?

If your agents call paid APIs, Locus Pro Enterprise turns each call into a charge to the right customer. Name the customer with X-Locus-End-User. The call draws that customer's balance at the catalog base price plus your markup, and your margin is recorded in the ledger.

Example research task with catalog prices from November 10, 2026: one Firecrawl search at 1.7 credits, three Firecrawl scrapes at a typical 1 credit each, and one Apollo person enrichment at a typical 35.4 credits. Base cost is 40.1 credits, or $0.0401. With a 30% markup the customer pays $0.05213. Scrape and enrichment prices vary by request, so a call can carry a price ceiling.

Limits keep agent costs bounded. An agent key can carry a credit limit per call and per loop, and each customer can have a monthly budget and a maximum charge per call. A call over a budget or ceiling is refused before it runs.

How do I choose a pricing model for my agent?

Answer three questions in order. Then test the answer against real usage data: your top 10% of customers by usage decide whether a flat price holds.

Can the buyer count the unit without your help? If not, rule it out.
Does your cost move with the unit? If cost moves and price does not, add an allowance and overage.
Who should pay for a runaway task? If you, cap the cost per task. If the buyer, give them spend limits.

Frequently asked questions

Should an AI agent be priced like software or like labor?

Price it by what the buyer counts. Agents that complete a task a person used to do, such as resolving tickets, can price per result. Agents that help people work faster usually price by seat with usage on top.

How much does Salesforce charge for Agentforce?

Salesforce lists $2 per conversation, or Flex Credits at $500 per 100,000 credits, where one action in its examples is 20 credits, or $0.10. The Agentforce User License is $5 a user a month and requires Flex Credits.

How do I stop an agent from running up costs on a fixed price?

Cap the cost of each task and each customer. Set a maximum number of steps or a credit limit per run, and send easy steps to cheaper models.

Can I change pricing models later?

Yes, but a new model changes customer bills. Before you switch, show customers their past usage priced under the new model, and give them a transition period.

Implementation references

Use these first-party references for current request contracts and account requirements. Tool availability, prices, and negotiated terms can change.

Sources

Third-party details were checked on November 10, 2026. Vendors change pricing and features often, so confirm on their site before you decide.