AI agent pricing is the Wild West of the software industry. Some platforms charge per subscription. Others bill per resolution, per execution, per token, or per operation. Some are free to install but rack up hundreds in monthly LLM API costs. Others charge a flat fee that includes everything — until you hit a limit nobody mentioned during sign-up.
The result: two teams building similar agents can end up paying anywhere from $20/month to $2,000/month depending on which platform they chose, which models they use, and how many conversations their agents handle. This guide breaks down what you actually pay across every major category — no-code platforms, low-code workflow tools, and developer frameworks — including the hidden costs that most pricing pages conveniently omit.
Master Pricing Table
| Platform | Free Tier | Starter Plan | Pro / Team Plan | Enterprise | Billing Model |
|---|---|---|---|---|---|
| Lindy | 400 credits/month | — | Pro: $49.99/month (5,000+ credits) | Business: $299.99/month | Credit-based (variable per task complexity) |
| Gumloop | — | $37/month | $97/month | Custom | Subscription + usage tiers |
| Zapier Agents | Limited tasks | From $20/month | Team: from $69/month | Custom | Per-task (agent features on higher plans) |
| Botpress | Limited bots | $45/month | Custom | Custom | Subscription + per-conversation |
| n8n | Free (self-hosted) | $24/month (cloud) | $60/month | Custom | Flat subscription (no per-operation charge) |
| Make.com | 1,000 ops/month | $10.59/month | $18.82/month | Custom | Per-operation |
| LangChain / LangSmith | Free (open source) | LangSmith: $0 (dev) | $39/seat/month | Custom | Open source + observability subscription |
| CrewAI | 50 executions/month | — | $25/month (100 executions) | Custom | Open source + per-execution (cloud) |
| AutoGen | Free (open source) | — | — | — | Open source (you pay infrastructure) |
No-Code Platform Pricing
Lindy uses a credit-based system that makes budgeting tricky. Simple tasks (sending a Slack message, checking your calendar) consume roughly 1 credit. Complex tasks (web research, data extraction, multi-step reasoning) consume 5–10+ credits. The free tier’s 400 monthly credits is enough to evaluate the platform over a few days. The Pro plan at $49.99/month includes 5,000+ credits, which covers moderate daily use for a single professional. Additional credits cost $10 per 1,000. Voice calls via Gaia start at $0.19/minute. For teams, Business at $299.99/month provides 30,000 credits with admin controls, SOC 2 compliance, and HIPAA support.
Gumloop charges a flat subscription starting at $37/month for individuals. The $97/month team plan increases workflow limits and adds collaboration features. Pricing is more predictable than credit-based systems because the subscription covers a defined number of workflow executions rather than variable-cost credits. For teams doing heavy pipeline work — lead qualification, content creation, data enrichment — the predictability is a genuine advantage over Lindy’s credit model.
Zapier Agents are available on paid Zapier plans starting at $20/month. Agent features — autonomous reasoning, web browsing, and multi-step task execution — require Team plans ($69/month) or higher for meaningful use. The underlying Zap infrastructure uses per-task pricing that becomes expensive at scale: a workflow with 10 steps counts as 10 tasks. Teams already on Zapier can add agents incrementally. Teams starting from scratch may find Lindy or Gumloop more cost-effective for agent-specific work.
Botpress starts free with limited bots and conversations. The Pro plan at $45/month unlocks higher conversation volumes and advanced features. For customer-facing chatbots and voice agents, Botpress’s per-conversation pricing model aligns costs directly with business value — you pay more as your bot handles more customers, which typically corresponds to growing revenue.
Low-Code / Automation Platform Pricing
n8n is the most cost-effective option for technically comfortable teams. The self-hosted version is completely free with no per-operation charges — the full feature set, including AI nodes, runs on your own infrastructure. The cloud-hosted option starts at $24/month. At scale, the cost advantage over per-operation platforms is dramatic: businesses switching from Zapier to n8n self-hosted report cutting automation costs by 70–90%, primarily because n8n eliminates per-operation pricing entirely. The trade-off is that self-hosting requires server management. Cloud hosting removes this burden for a predictable monthly fee.
Make.com charges per operation, with the free tier covering 1,000 operations per month and paid plans starting at $10.59/month. AI-enhanced steps count as operations. The risk is that AI workflows generate more operations than traditional automations — an agent that classifies an email, looks up the sender in a CRM, drafts a reply, and logs the interaction consumes at least 4 operations per email. At 50 emails per day, that’s 6,000 operations per month — requiring a higher plan tier than the volume might suggest. Make is excellent value for light automation but can become expensive when AI agent workflows generate high operation counts.
Developer Framework Costs
Developer frameworks are free to install — but the true cost of ownership extends far beyond the framework licence.
LangChain / LangGraph is free and open-source. The real costs are LangSmith (observability and monitoring, free for developers, $39/seat/month for teams), LLM API calls (the dominant expense), and hosting infrastructure. A production agent built on LangChain typically costs: $0 for the framework + $39–200/month for LangSmith + $50–500/month in LLM API costs + $20–200/month for hosting (cloud VMs, vector databases, queues). Total range for a single production agent: $100–900/month depending on model choices and traffic volume.
CrewAI follows a similar pattern but with clearer managed options. The open-source core is free. CrewAI Cloud offers a free tier (50 executions/month), Professional at $25/month (100 executions), and custom Enterprise pricing with self-hosted Kubernetes deployment. Execution-based pricing means complex multi-agent workflows with many steps hit limits quickly. A three-agent crew doing 20 research tasks per day would exhaust the Professional plan’s 100 monthly executions in five days. For production workloads, budget for Enterprise pricing plus LLM API costs.
AutoGen (Microsoft) is entirely free and open-source. You pay only for infrastructure and LLM API calls. This makes it the cheapest entry point for developers who can manage their own hosting. The trade-off is no managed monitoring, no cloud dashboard, and no commercial support — you’re responsible for everything.
Sample monthly costs for a developer-built agent at different scales:
| Scale | LLM API | Infrastructure | Monitoring | Framework | Total |
|---|---|---|---|---|---|
| Prototype (100 tasks/day) | $30–80 | $20–40 | $0 (free tier) | $0 | $50–120 |
| Production (1,000 tasks/day) | $200–600 | $50–150 | $39–100 | $0–25 | $290–875 |
| Enterprise (10,000 tasks/day) | $1,000–5,000 | $200–800 | $200–500 | Custom | $1,400–6,300 |
The Hidden Costs
Most agent pricing pages show the platform subscription. What they don’t prominently feature are the costs that typically exceed the platform fee.
LLM API calls are the single largest hidden cost. Every time an agent reasons about a decision, it calls a language model. A customer support agent handling 100 conversations per day, averaging 1,500 tokens per conversation (a modest exchange), consumes roughly 4.5 million tokens monthly. At Claude Sonnet 4.6’s pricing ($3/$15 per million tokens), that’s approximately $80–100/month in token costs alone — on top of whatever platform fee you’re paying. Using Opus 4.6 ($15/$75) for the same workload would cost $400–500/month. Using DeepSeek V3.2 ($0.28/$1.10) brings it under $10/month. Model selection is the single most impactful cost lever you control.
Tool and integration API costs add up quietly. Each API your agent calls — CRM lookups, email sends, web searches, database queries — may carry its own per-call charge. An agent that researches leads by querying LinkedIn, checking company websites, and enriching data via Clearbit can accumulate $0.10–0.50 per lead in third-party API costs before the agent’s own LLM usage.
Compute and hosting costs for developer-built agents include cloud VMs ($20–200/month), vector databases for RAG ($20–100/month), message queues for reliable task processing ($10–50/month), and storage for conversation histories and logs. These are invisible on no-code platforms (included in the subscription) but significant for self-hosted deployments.
Monitoring and logging at scale becomes its own line item. LangSmith, Helicone, or custom observability stacks cost $39–500/month for production deployments. Without monitoring, you won’t know when your agent is making bad decisions or burning through tokens on reasoning loops — which means the absence of monitoring is itself a hidden cost in the form of undetected waste.
A realistic total budget multiplier: industry analysis suggests that the true total cost of an AI agent deployment is approximately 1.7× the base token calculation. Factor in growth (+25%), infrastructure overhead (+30%), and experimentation budget (+15%) for a realistic annual figure.
Total Cost of Ownership by Scale
| Team Size | Agent Count | Typical Setup | Monthly Cost Range | Notes |
|---|---|---|---|---|
| Solopreneur | 1–3 agents | Lindy Pro or n8n self-hosted + cheap LLM | $20–100/month | Credit-based platforms work well; DeepSeek or Gemini Flash keep API costs low |
| Small business (5–20 employees) | 5–10 agents | Gumloop or Zapier Agents + GPT-5.4 / Sonnet | $150–600/month | Platform fees + LLM costs per agent; monitor per-agent usage carefully |
| Mid-market (20–100 employees) | 10–30 agents | n8n self-hosted or CrewAI Cloud + model routing | $500–3,000/month | Self-hosting saves on platform fees; model routing (cheap model for simple tasks, frontier for complex) controls API costs |
| Enterprise (100+) | 50+ agents | Custom framework (LangChain/LangGraph) + managed infrastructure | $3,000–15,000+/month | LLM API costs dominate; negotiate volume pricing with providers; dedicated observability infrastructure |
The most common budget mistake: underestimating LLM API costs. Platform subscriptions are predictable. Token consumption is not. Set spending limits on your API accounts from day one, and implement model routing — use cheap models for simple decisions and reserve expensive frontier models for tasks that genuinely require deep reasoning.
Frequently Asked Questions
What’s the cheapest way to start with AI agents?
Three paths, depending on your technical comfort. For non-technical users: Lindy’s free tier (400 credits/month) or Make.com’s free tier (1,000 operations) costs $0 and provides enough to build and test a simple agent. For slightly technical users: n8n self-hosted is completely free with the full feature set — you just need a basic server ($5–10/month on a cloud provider). For developers: AutoGen or CrewAI’s open-source core plus a DeepSeek API key ($3–5/month for light usage) gets you a functional multi-agent system for under $10/month. The key is starting with a cheap LLM (DeepSeek, Gemini Flash) and upgrading to frontier models only for tasks that genuinely need the quality.
Why do agent costs vary so much?
Three variables drive most of the variation. First, model choice: Claude Opus 4.6 costs 50× more per token than DeepSeek V3.2. An identical agent doing identical work costs dramatically different amounts depending on which model powers it. Second, conversation volume: an agent handling 10 tasks per day costs a fraction of one handling 1,000. Third, task complexity: a simple email classifier consumes a few hundred tokens per decision. A multi-step research agent might consume 50,000+ tokens per task, including web browsing, document analysis, and report generation. The platform subscription is typically the smallest portion of the total bill — model costs and usage volume determine the real number.
Are per-resolution or per-execution models better?
Per-resolution pricing (pay per completed customer interaction) aligns costs with business value — you pay more as the agent handles more customers, which should correspond to revenue growth. Per-execution pricing (pay per workflow run) is more predictable for internal operations where “value per run” is harder to measure. For customer-facing agents, per-resolution is generally preferable because it creates natural cost-to-value alignment. For internal automation, flat subscriptions or self-hosted options (n8n, open-source frameworks) provide better cost predictability. The worst model for budget planning is credit-based billing where credit consumption varies unpredictably by task complexity.
Read next:
- Best AI Agent Platforms in 2026: The Complete Comparison
- Best AI Agent Builders for Non-Technical Users
- Agent Frameworks for Developers: LangChain vs CrewAI vs AutoGen
- Zapier vs n8n vs Make: Which Automation Platform?
AI Agent Brief is editorially independent. Our recommendations are based on hands-on testing, not advertising relationships. When you subscribe to a tool through our links, we may earn a commission at no extra cost to you. This never influences our rankings.
© 2026 AI Agent Brief. All rights reserved.
Also in this series