AI agent pricing in 2026: credits, seats and per-resolution fees explained

Published by 3Gen Internet Corporation, maker of OpenAssistant, Skej and AGNT. Comparisons use public documentation, without hands-on testing.

AI agent pricing falls into six patterns: per user per month with a credit tier, credits, tasks or actions, per resolution, flat subscriptions with usage limits, and enterprise platform fees. OpenAssistant uses the first one: a price per user per month that includes a monthly credit allowance, plus a free Starter plan. Before you compare any two prices, find out what the unit is (a seat, a credit, a task, a resolution), what one unit buys, and what happens when you run out. Every price below comes from the vendor’s own page and can change.

Pattern 1: per user per month with a credit tier (OpenAssistant)

You pay a flat price for each person, and each person gets a monthly credit allowance. OpenAssistant works this way. Pro is $72 per user per month billed monthly, or $60 on the annual toggle. Team is $120 monthly or $100 annual. Both prices are for the 5,000-credit tier (OpenAssistant pricing; per-user pricing confirmed by 3Gen).

The details that decide what you really pay:

• A free plan exists. Starter gives 1,000 credits at signup, then 100 bonus credits a day once the balance is used up, up to 1,000 a month (OpenAssistant pricing). Bonus credits are only granted when the balance is used up (pricing details).
• Credits do not roll over. The vendor says plan credits reset each billing cycle, do not roll over and are not refundable. Purchased top-up credits do not expire while the subscription is active (pricing details; credits help article).
• A credit buys a variable amount of work. Simple tasks might use 10 to 50 credits; deep research and complex recurring tasks use hundreds (pricing details). The help center adds that there is not always a fixed credit cost for a request, and that task effort (Low, Medium or High) can change the credits used (credits help article).
• Higher credit tiers cost more. The pricing details page lists credit tiers from 5,000 to 100,000 a month for Pro and Team, with price scaling by tier. The rendered pricing page shows only the 5,000 tier, so we do not state prices for the others.
• Team seats start billing at invite. Billing for a new seat begins when the invitation is sent, and an admin who raises a user’s credit allocation raises that seat’s price by the same multiplier (team billing help article).

Where a competitor does better: OpenAssistant is a general work assistant, not a support-desk product. If your job is resolving customer tickets, a per-resolution price (Pattern 4) tells you the cost of each solved ticket in advance, which a credit price does not.

Pattern 2: credits

Credits are a shared currency that different actions consume at different rates. The vendor sets the exchange rate, so the same number of credits can buy very different amounts of work.

• Lindy sells credits per active user. The 5,000-credit Team tier is $49.99 per active user per month. Some public versions also list $29.99 for 3,000 credits, while the live browser dropdown begins at 5,000 credits; check available tiers at signup. The page has a slider up to 1,750,000 credits at $9,999.50, and says every active user in the workspace is billed. Examples on the page: everyday asks use 2 to 250 credits, deep work 250 to 1,000 and big builds 1,000 to 2,500 (Lindy pricing). The docs price base-plan credits at $0.01 each and say monthly credits do not roll over, while purchased top-ups do (Lindy docs). At $0.01 a credit, an everyday ask costs between $0.02 (2 x $0.01) and $2.50 (250 x $0.01). The Free plan gives “$50 in credits for free” that last 7 days.
• Make bills in credits, with most actions using 1 credit. At 10,000 credits a month, Core is $16 monthly or $12 on annual billing, and credits expire at the end of the term (Make pricing). Free is 1,000 credits a month.
• Manus lists three paid cards at $20, $40 and $200 a month on the monthly toggle, with 4,000, 8,000 and 40,000 credits (Manus pricing). The page does not show plan names. The docs say plan credits reset monthly and purchased add-on credits never expire (Manus docs).
• Perplexity Computer uses credits inside Perplexity subscriptions. The help center says “100 credits equals $1”, lists Max with 10,000 credits a month, and gives task examples from 15 to 70 credits for a quick fact up to 350 to 950 credits for a research report (Perplexity credits help).

Pattern 3: tasks, actions and executions

Automation platforms that added AI agents often kept their old unit. You pay per task, activity, action or workflow run, and the AI step may count for more than one.

• Zapier has two views. On the pricing page, Agents Free is 400 activities a month and Agents Pro is $400 billed annually for 1,500 activities. The help center defines an activity as any billable action the agent takes (Zapier pricing, activity help article). On the platform plans, AI steps are billed in tasks: Free is 100 tasks a month and Professional starts from $19.99 a month. AI by Zapier steps cost 1, 3 or 5 tasks depending on model tier, and each tool call is billed separately (Zapier rates).
• n8n counts workflow executions, “regardless of complexity”, and every plan has unlimited users and workflows. Starter is 20 EUR a month billed annually for 2.5K executions and Pro is 50 EUR for 10K (n8n pricing).
• Relevance AI counts Actions, which are counted when an agent runs a tool, plus Vendor Credits for model costs passed through at wholesale. Docs list Pro at $19 a month annual or $29 monthly with 2,500 Actions a month, and Team at $234 annual or $349 monthly with 7,000 Actions a month. Action top-ups are $80 per 1,000 (Relevance AI docs). The public pricing page shows no figures and says to talk to sales.

Pattern 4: per resolution or per outcome

Support agents often charge only when the agent finishes the job. The bill tracks results, but you need to know what counts as one.

• Intercom Fin charges $0.99 for service resolutions and configured Procedure handoffs; sales qualifications cost $9.99. A simple pass-through without an outcome is uncharged, and only one billable outcome is charged per conversation. Existing-helpdesk use lists a 50-outcome monthly minimum (about $49.50) and no extra seat costs (Fin pricing). Intercom Essential displays a $19-per-seat monthly new-customer offer against $29 list; confirm billing and eligibility (Intercom pricing).
• Zendesk charges seats plus resolutions. Support Team is $19 per agent per month paid yearly, or $25 monthly. The page says you pay only for requests the AI agent resolved, with 5 included resolutions per agent per month on Support Team, then $1.50 each committed or $2.00 each pay-as-you-go (Zendesk pricing). Zendesk says it introduced resolution tiers on May 18, 2026, and that an LLM runs a verification step on resolutions (Zendesk help article). The dollar value of each tier is not stated on the pages we read.

For a deeper look at the support tools themselves, see best AI customer support agents.

Pattern 5: flat subscriptions with usage limits

Chat-style agents usually sit inside a monthly subscription. The price is flat, and the limit is a usage allowance the vendor rarely states in numbers.

• ChatGPT lists Free at $0, Go at $8 a month (the page says it may include ads), Plus at $20 and Pro from $100 (ChatGPT pricing). The help center lists three Pro tiers: $100, $200 and $500, billed monthly. It says new Pro 200 subscriptions get a lower usage allowance at an unchanged $200 price (Pro tiers help). The Business tab lists a standard seat at $20 a month billed annually or $25 monthly, and a premium seat at $100 annual or $125 monthly with 5x the standard usage. The old “ChatGPT agent” help article now says it is no longer available and points to ChatGPT Work, and that page gives no per-plan Work usage numbers (ChatGPT agent help).
• Claude lists Free at $0, Pro at $17 a month on the annual discount ($200 billed up front) or $20 monthly, and Max from $100 a month at 5x or 20x Pro usage. Team standard seats are $20 a month annual or $25 monthly, premium seats $100 or $125. Enterprise is $20 per seat per month billed annually with a 20 seat minimum, plus usage at API rates. The page footnote says usage limits apply, and prices can change at Anthropic’s discretion (Claude pricing).
• Perplexity lists Pro at $17 a month billed annually and Max at $167 a month billed annually (Perplexity plans). The help center says Max costs $200 monthly or $2,000 annually (Perplexity Max help). The Pro monthly price is not stated on the pages we read.
• Genspark mixes both models: Plus is $24.99 a month, or $19.99 a month billed annually, with credits from 10,000 a month. Team is $40 per seat per month, or $35 billed annually (Genspark plans).

For free tiers across these products, see best free AI agents. For coding tools, see AI coding agent pricing.

Pattern 6: enterprise platform fees

Large platforms combine a licence, a consumption meter and a cloud account. These prices are for building and running agents, not for one person’s assistant.

• Salesforce Agentforce lists Flex Credits at $500 per 100,000 credits. An Agentforce action costs 20 Flex Credits (20 / 100,000 x $500 = $0.10) and a voice action 30. Conversations are $2 each. Add-ons for Sales, Service and Field Service are $125 per user per month for unmetered employee usage. The page says prices are subject to change and to contact sales for details (Agentforce pricing).
• Microsoft Copilot Studio sells capacity packs of 25,000 Copilot Credits at $200 per pack per month, tenant-wide. It also offers pay-as-you-go and a pre-purchase plan with savings up to 20%, and an Azure subscription is required to use agents (Copilot Studio pricing). Credit rates are 1 for a classic answer, 2 for a generative answer and 5 for an agent action (Microsoft credit rates).

For builder platforms side by side, see best AI agent builders.

Published prices in one table

Listed by pricing pattern, not ranked. Prices are USD unless noted, and each product name links to the page the figures come from. OpenAssistant rows are per user per month (per-user pricing confirmed by 3Gen).

Worked example: a 10-person team

All figures come from the published prices above. The arithmetic shows list prices only, and the products do not include the same things, so this is a way to see the shape of each bill, not a verdict on value.

A. Ten people each get an assistant or chat agent, per month

The OpenAssistant rows rest on (OpenAssistant pricing; per-user pricing confirmed by 3Gen). Each OpenAssistant seat includes 5,000 credits a month, so ten Pro seats hold 10 x 5,000 = 50,000 credits, and unused plan credits do not roll over. At 10 to 50 credits for a simple task, 5,000 credits cover 5,000 / 50 = 100 to 5,000 / 10 = 500 simple tasks per person per month. Annual-toggle totals over a year: Pro 10 x $60 x 12 = $7,200 and Team 10 x $100 x 12 = $12,000. The vendor describes Pro as meant for individual users and small startup teams and Team for larger organizations (pricing details). Perplexity’s $40 is the “starts at” Enterprise Pro seat price in its plan comparison help article , and the Genspark and ChatGPT figures come from the pages linked above. The seat rows buy different things: a credit allowance in one case, a usage allowance the vendor does not number in the others.

B. A 10-agent support team with 200 AI resolutions in a month

Notes: Zendesk seats use the yearly price; at the $25 monthly price the seat line is 10 x $25 = $250. The Zendesk allowance and the $1.50 and $2.00 rates come from the pricing page, and the help center says tier dollar values are not shown there. Fin’s page does not spell out whether the first 50 outcomes inside the 50-outcome minimum are billed separately from the rest, so $198 is the per-outcome figure alone. Agentforce is priced for customer-facing agents by Flex Credits or Conversations, not by the $125 add-on. For action-based billing, 1,000 Agentforce actions use 1,000 x 20 = 20,000 Flex Credits, which is 20,000 / 100,000 x $500 = $100 (Fin pricing, Zendesk pricing, Agentforce pricing).

C. Credit packs translated into actions (Copilot Studio)

One pack is $200 for 25,000 credits. At 5 credits per agent action that is 25,000 / 5 = 5,000 actions, or $200 / 5,000 = $0.04 each. At 2 credits per generative answer it is 25,000 / 2 = 12,500 answers, or $200 / 12,500 = $0.016 each (Copilot Studio pricing, credit rates).

Questions to ask about the bill

1. What is the billing unit, and what does one unit buy for my kind of work? (A credit can be 2 or 250 credits of work.)
2. Who counts as a user? Does an invited but inactive person get billed?
3. Do unused credits or allowances roll over, and do top-ups behave differently?
4. What happens at the limit: does the agent pause, stop, or bill overage at a higher rate?
5. Do model choice, tool calls or voice change how many units a step uses?
6. What counts as a resolution or outcome, who verifies it, and what is refunded?
7. Are there minimums, seat requirements or a separate cloud account?
8. Which billing view is the price shown for, annual or monthly, and which usage limits apply to the plan?

Hidden-cost traps the vendor pages record

• Allowances that expire. Make says credits expire at the end of the term. Copilot Studio says unused credits do not carry over to the next month. Lindy monthly credits and OpenAssistant plan credits do not roll over, while their purchased top-ups behave differently (Make, Copilot Studio billing). Zendesk’s older resolution platform expired allowances annually (Zendesk help).
• Multipliers inside a step. Zapier bills Premium-tier AI steps at 5 tasks a step, and each tool call is billed separately (Zapier rates). Make’s code app uses 2 credits per second of execution (Make). Copilot Studio adds a premium tools rate on top of the feature rate for reasoning models (Microsoft credit rates).
• Overage that keeps running. Zapier can bill extra tasks at a higher per-task rate if you enable it (Zapier pricing). Zendesk’s older platform allowed overage by default, with a pause option (Zendesk help). n8n Business overage is 4,000 EUR per extra bucket of 300,000 executions (n8n pricing). Copilot Studio custom agents are disabled when a tenant reaches 125% of prepaid capacity (Microsoft credit rates). By contrast, Lindy says it pauses and tells you when credits run out (Lindy pricing).
• Minimums and prerequisites. Fin has a minimum of 50 outcomes a month (about $49.50) on an existing helpdesk (Fin pricing). Claude Enterprise has a 20 seat minimum plus usage at API rates (Claude pricing). Copilot Studio needs an Azure subscription, and sharing agents with outside channels needs a standalone plan (Copilot Studio pricing). The Agentforce User License at $5 per user per month requires Flex Credits (Agentforce pricing).
• Seat counting. Lindy bills every active user. OpenAssistant Team seats bill from the moment an invite is sent (team billing help article).
• Allowances that change. ChatGPT Pro 200 new subscriptions get a lower usage allowance at the same price, and Claude says prices and plans are subject to change (Pro tiers help, Claude pricing). Perplexity’s bonus credits on Max expire in about 30 days (Perplexity credits help).
• Which price you are looking at. Several pages load on the annual view: OpenAssistant shows $60 for Pro until you switch to monthly ($72), and Perplexity’s $167 for Max is a billed-annually figure against $200 monthly.

Related reading

• AI coding agent pricing in 2026: what Claude Code, Codex, Devin and Copilot cost: Plans and prices for Claude Code, Codex, Devin, Copilot, Cursor, Replit and Jules, the stated limits, traps and a 5-developer example.
• Best AI customer support agents in 2026: prices per resolution and per seat: Seven support agents compared, with per-resolution and per-seat prices explained.
• Best free AI agents in 2026: what you get at no cost: True free plans, trials and expiring credits, with the stated limits and what each free tier lacks.
• Best AI agents for small business in 2026: top picks and prices: Eight agents for a 5 to 50 person business, with list prices, a ten-person seat cost and what each will not do.
• How to choose an AI agent: 10 questions to ask before you buy: Ten questions on job fit, channels, approvals, price, data and exit, with a job-to-product table.

Frequently asked questions

How do AI agents charge for usage?

Six ways appear on the pages we read: a price per user with a credit allowance, credits, tasks or actions, per resolution or outcome, flat subscriptions with usage limits, and enterprise platform fees that combine licences with consumption meters. Many products mix two, such as Zendesk (seats plus resolutions) and Genspark (subscription plus credits).

Is there a free plan for an AI agent?

Several vendors publish one. OpenAssistant Starter gives 1,000 credits at signup and 100 bonus credits a day once used up, up to 1,000 a month. Lindy’s free credits last 7 days, Zapier Agents Free is 400 activities a month, and Make Free is 1,000 credits a month. See best free AI agents for the full list.

Is per-seat or per-resolution pricing cheaper?

It depends on volume and the job. In the worked example above, 200 AI resolutions cost $198 at Fin’s per-outcome rate, while ten OpenAssistant Pro seats cost $600 to $720 a month (per-user pricing confirmed by 3Gen). Those two do different jobs: one resolves support tickets and the other is a general assistant for ten people. Compare cost for the job you need, not the headline unit.

Do unused credits roll over?

Usually not on the plan allowance. OpenAssistant, Lindy, Make, Copilot Studio and Relevance AI plan allowances reset or expire, while Manus add-on credits and OpenAssistant top-ups do not expire while the subscription is active. Check the specific rule for the plan you buy.

Try OpenAssistant

If you want an assistant that acts across email, text, phone calls and Slack for a flat per-user price, OpenAssistant has a free Starter plan with no card required. See the current plans and check the credit mechanics against your own workload before you pay for seats.