All articles
AI Assistant·

Make AI Agents Pricing for Small Business (2026): Credits, Tokens, and Real Cost

Make AI Agents uses Make credits rather than a separate public per-agent price. This vendor-independent guide explains operation and token costs, realistic SMB usage scenarios, hidden costs, and alternatives.

Make AI Agents does not have a separate public per-agent subscription in the documentation verified for this guide. Agent use draws from your Make plan's credits. With Make's AI Provider, a run costs at least one credit per operation plus a variable number of credits for AI tokens. With a custom OpenAI, Anthropic, or other provider connection, Make generally charges one credit per operation while the model provider bills tokens separately.

That means the real monthly cost is: Make plan + agent/tool operations + AI-token usage + connected-app fees + setup and review labour. A small, tightly scoped agent may fit inside an existing credit allowance. A chatty agent with several tools, large files, or repeated record processing can consume far more than an ordinary fixed scenario.

This guide was checked against Make's Help Center on August 16, 2026. It is vendor-document-verified editorial analysis, not a hands-on test. Make's public pricing page returned a Cloudflare verification screen in this research environment, so current self-serve dollar prices could not be independently read from that page. We do not substitute stale third-party prices.

Diagram showing the components of a Make AI Agents monthly cost

Agent cost is a stack, not one line item. The plan is only the bottom layer.

Make AI Agents pricing at a glance

The safest answer is to budget in credits first, then map that demand to the live Make plan selector in your region.

Cost componentMake's AI ProviderCustom AI provider connection
Run an agent1 credit per operation + credits based on AI tokens1 Make credit per operation
Agent chat1 credit per operation + tool-operation credits + AI-token credits1 credit per operation + tool-operation credits; provider bills model tokens
Agent tools1 credit per operation, plus AI-token credits when applicable1 Make credit per operation; provider token charges may apply
Knowledge uploadOperation credit plus embedding/file-description token usage; PDF/DOCX also includes 10 tokens per pageOperation credit plus embedding usage; provider may bill applicable tokens
Input filesToken usage varies with file size, type, model, and providerProvider token usage varies with file size, type, and model
AvailabilityMake's AI Provider is documented for all plansCustom provider connections are documented for paid plans

Make's Help Center says Core supports up to 300,000 monthly credits and Pro up to 8 million. Make also documents a 25% premium for manually purchased or automatically purchased extra credits. Those ceilings are not entry prices: verify the current plan, included credits, billing cadence, currency, tax, and regional offer on Make's pricing page before buying.

What does one Make credit pay for?

For most non-AI modules, one operation equals one credit. A trigger check costs a credit even when it finds nothing. A search module normally costs one credit per run, but downstream actions cost one credit for every bundle they process. Routers and filters themselves are free, while iterators can multiply the paid work after them.

AI changes the arithmetic. Make distinguishes fixed credit use from dynamic credit use:

  • Fixed: a known number of credits per module operation.
  • Dynamic: credits vary with tokens, pages, file size, processing time, or another measured factor.
  • Make's AI Provider: Make converts model input and output tokens into credits and adds operation credits.
  • Custom provider: Make charges operation credits; the connected model provider charges its own token bill.

Make's current documentation lists its Small, Medium, and Large model tiers at 5,000, 3,500, and 1,500 tokens per credit respectively. Other named models use different input and output conversion rates, and Make warns that models and rates can change. Treat the in-product usage record—not a spreadsheet made once—as the final meter.

How is agent usage different from an ordinary Make scenario?

An ordinary scenario follows a path you designed. An agent can interpret a request, choose tools, consult knowledge, and decide what action to take next. That flexibility creates more variable usage.

Consider a deterministic lead workflow: one trigger, one CRM search, one update, and one notification. It is usually straightforward to estimate. An agent handling the same lead may read instructions, call a search tool, consult knowledge, retry a tool, generate text, update the CRM, and ask for approval. Each operation and the AI tokens can contribute to cost.

Agent chat also counts tool calls separately. A conversation that looks like one user request may create several billable platform operations. Longer prompts, long chat history, large input files, verbose output, and unnecessary knowledge retrieval raise token demand.

Use a standard scenario when the logic is known. Use an agent only where interpretation or tool selection genuinely earns its keep. Putting an AI agent around a predictable three-step process is usually automation wearing an expensive hat.

A dated cost model for light, moderate, and high usage

The following model is a planning tool dated August 16, 2026, not a Make quote or benchmark. Replace every assumption with usage from a controlled pilot.

Usage levelEditorial workload assumptionEstimated Make demandSeparate model-provider costOperational cost
Light100 agent runs/month; 3 tool operations/run; short promptsRoughly 400+ operation credits, plus token credits if using Make's providerOften $0 with Make's provider; roughly $0-$10 with a custom provider, model-dependent2-5 setup hours, then 1-2 review hours/month
Moderate1,000 runs/month; 5 tool operations/run; some knowledge useRoughly 6,000+ operation credits, plus agent and embedding tokensRoughly $5-$75/month is a reasonable test budget, not a promise10-30 setup hours, then 4-10 review hours/month
High volume10,000 runs/month; 7 tool operations/run; files, retrieval, approvalsRoughly 80,000+ operation credits before variable token use and retriesRoughly $50-$500+/month depending on model, context, output, and retry rate40-120+ setup hours and ongoing ownership

These ranges intentionally separate provider spend from Make spend. The provider range is an editorial contingency allowance because model prices and prompts vary dramatically; it is not a forecast from Make. With Make's AI Provider, token cost appears inside Make credits instead.

Three planning bands for Make AI Agents usage

Start with measured operations per successful outcome, then add token usage, retries, and headroom.

For a defensible estimate, log 50 to 100 representative runs and calculate:

  1. average agent operations per completed outcome;
  2. average called-tool operations;
  3. input, output, and embedding tokens;
  4. retry and failure rate;
  5. peak-day volume; and
  6. human review minutes per outcome.

Add 20% to 30% headroom. If your plan depends on every run behaving like the clean demo, it is not a plan. It is fan fiction with a calculator.

What might 5-, 10-, and 25-person teams pay?

Headcount is a poor usage unit. One busy support queue can cost more than 25 occasional users, so these scenarios use workload—not seats—as the driver.

TeamSensible starting designMonthly usage assumptionBudget posture
5 peopleOne builder, one narrow agent, approval before external actions100-300 runs; about 400-2,000 operation credits plus tokensStart inside the smallest suitable live plan; reserve $0-$15 for separate provider tokens and price 2-5 hours of setup
10 peopleTwo or three agents for sales/support triage with shared review800-2,000 runs; about 5,000-15,000 operation credits plus tokensSelect a plan with at least 25% headroom; reserve $10-$100 for provider tokens and 4-10 monthly governance hours
25 peopleMultiple production agents, knowledge, role controls, approvals5,000-15,000 runs; about 40,000-150,000+ operation credits plus tokensCompare Pro/Teams/Enterprise entitlements, support, roles, and region; reserve $50-$500+ for provider tokens and a named owner

The table does not state a universal subscription total because Make's current dollar pricing surface could not be read in this verification environment. Before approval, paste the live plan quote into the model and record the currency, cadence, included credits, extra-credit rate, tax, and renewal terms.

Which plan and team entitlements need verification?

Make documentation confirms that Make's AI Provider is available on all plans and custom provider connections are available on paid plans. It does not make every collaboration or governance entitlement universal.

Before buying, confirm:

  • how many users, teams, and organizations the plan permits;
  • whether custom roles, team roles, or granular permissions are included;
  • whether audit logs, SSO, advanced security, and priority support require Enterprise;
  • who may create, edit, run, or inspect agents;
  • whether credit pools are shared and who receives usage alerts;
  • which AI models and provider connections are available in your region;
  • data-processing location, retention, subprocessors, and contract options; and
  • any execution-time, scenario-frequency, concurrency, file-size, or knowledge limits shown in the current account.

Plan names are not governance controls. A 25-person company should buy the entitlement that creates clean ownership and review boundaries, not simply the cheapest credit bucket.

Hidden costs that do not appear in the headline plan

  • Extra credits: Make documents a 25% premium over included-credit pricing for manual and automatic extra-credit purchases.
  • Model-provider charges: custom OpenAI, Anthropic, Gemini, or other connections generate a separate bill.
  • Connected software: enrichment, email, CRM, database, search, and document tools may have usage charges.
  • Knowledge ingestion: uploads, embeddings, page processing, and file descriptions can consume tokens or credits.
  • Polling: triggers can consume credits even when no new records are found.
  • Bundle fan-out: one search can return many records, multiplying downstream actions.
  • Retries and error handling: failed tool calls still create usage and investigation time.
  • Build and testing labour: prompts, permissions, schemas, filters, test cases, and exception paths need work.
  • Human review: approvals reduce risk but add time and queue management.
  • Security and procurement: SSO, legal review, data-processing terms, and regional requirements may push the buyer toward Enterprise.
  • Maintenance: app schemas, credentials, model behaviour, credit conversions, and knowledge files change.

What governance and human-approval questions matter?

Start with consequence, not novelty.

  • Can the agent send an external message, spend money, alter a contract, or delete data?
  • Which actions always require approval?
  • Are tool connections scoped to the minimum permissions?
  • Who owns the agent instructions, test suite, and incident response?
  • Can a reviewer see the evidence used for a decision?
  • What happens when knowledge is missing, stale, or contradictory?
  • Is sensitive data allowed to reach the selected model provider and region?
  • How are failed, partial, and duplicate actions detected?
  • What daily or monthly cost threshold pauses the workflow?
  • Can the business reproduce and reverse an action?

Human approval is sensible for payments, customer-facing commitments, deletions, employment decisions, legal documents, and high-impact record changes. “Autonomous” is not a maturity level. Sometimes it is just negligence with better branding.

Decision flow for automatic, approval-required, and blocked agent actions

Permission should follow consequence: automate low-risk actions, approve consequential ones, and block prohibited ones.

When is standard Make automation enough?

Use a regular Make scenario when inputs are structured and the decision can be expressed with filters, routers, mappings, and deterministic actions. It will usually be cheaper, easier to test, and easier to audit.

Use Make AI Agents when a workflow must interpret varied language, select among tools, combine instructions with knowledge, or handle exceptions that would otherwise require an unwieldy branch tree.

A practical architecture often combines both: an agent classifies or proposes; a deterministic scenario validates fields, enforces limits, requests approval, and performs the final action.

When might Zapier Agents or n8n fit better?

AlternativeBetter fit whenMain tradeoff
Zapier AgentsYour team already depends on Zapier, wants a simpler agent interface, and its activity allowance fits the workloadAgents uses a separate activity meter and may add cost beside the core Zapier plan
n8nYou want self-hosting, code-level control, or execution economics suited to high-volume technical workflowsMore hosting, security, observability, and engineering responsibility
Standard Make scenariosThe process is predictable and visual orchestration is the real needLess flexible with ambiguous or conversational inputs

Do not compare “credits,” “activities,” and “executions” as if they were equivalent. Model one real workflow in each platform, count all tool calls and downstream actions, include model tokens, and price the person who maintains it.

Is Make AI Agents worth it for a small business?

Make AI Agents is worth considering when the company already uses Make, the work needs interpretation, and the agent can be confined to a narrow outcome with measurable usage and approval controls. The product is less attractive when the workflow is deterministic, high-volume, difficult to audit, or dependent on broad permissions.

Run a small pilot, measure cost per successful outcome, and compare it with a standard scenario. Choose the agent only if flexibility produces enough value to justify variable credits, provider charges, and governance work.

Frequently asked questions

Is Make AI Agents free?

Make documents its own AI Provider as available on all plans, including Free, but use still consumes plan credits. Custom AI provider connections require a paid Make plan and generate a separate provider bill.

Are Make AI tokens charged separately?

With Make's AI Provider, token consumption is converted into Make credits. With a custom provider connection, Make charges operation credits and the AI provider bills its tokens separately.

Is Make AI Agent (New) still beta?

The current Help Center labels the product Make AI Agent (New) and maintains current setup, use-case, best-practice, and credit documentation. The pages checked did not display a beta warning or a general-availability declaration. Buyers should confirm release status and support commitments in their own region and contract rather than infer GA from the name.

Do unused Make credits roll over?

The sources checked for this guide did not establish rollover. Confirm the current plan terms before purchase and budget on a monthly allowance unless Make states otherwise in your quote.

Was first-party performance evidence available?

No Google Search Console or GA4 evidence was supplied. This article makes no traffic, conversion, productivity, or ROI claim.

Sources and verification

Product facts were checked August 16, 2026 against Make's Credits guide, Credit usage for AI agents, How features use credits, Make AI Agent (New), and November 2025 plan and pricing adjustments. The Make pricing page returned a Cloudflare verification screen in this environment, so buyers should verify current plan prices and regional terms directly. Vendor documentation establishes billing rules and availability claims; it does not independently establish business outcomes.