Understanding AI API costs: the numbers you need to know
The most opaque part of AI automation budgeting is API costs. "Pay per token" sounds either free or expensive depending on your expectations — the reality is neither. Let me make it concrete.
What is a token?
Tokens are the units that AI models use to process text. As a rough guide: one token is approximately 4 characters of English text, or about 0.75 words. A typical short email (subject + 200-word body) is approximately 300–400 tokens. A typical one-page document is approximately 750–1,000 tokens. The AI's response (output tokens) is charged separately and at a higher rate than input tokens in most models.
OpenAI model costs (verified November 2024)
OpenAI API pricing — what you actually pay per task
| Model | Input cost/1M tokens | Output cost/1M tokens | Cost per email (est.) | Cost per 1,000 emails |
|---|---|---|---|---|
| GPT-4o mini | $0.15 | $0.60 | ~$0.0001 | ~$0.10–$0.50 |
| GPT-4o | $5.00 | $15.00 | ~$0.003 | ~$3.00–$15.00 |
| GPT-3.5 Turbo | $0.50 | $1.50 | ~$0.0003 | ~$0.30–$1.50 |
| Claude 3.5 Sonnet | $3.00 | $15.00 | ~$0.002 | ~$2.00–$10.00 |
| Claude 3 Haiku | $0.25 | $1.25 | ~$0.0002 | ~$0.25–$1.25 |
| Whisper (audio) | $0.006/min | — | ~$0.06/meeting | ~$60/1,000 meetings |
Email estimates assume 300 input tokens (subject + body excerpt) + 150 output tokens (classification JSON + short draft). Actual costs vary significantly by prompt length and output length.
The most important takeaway from these numbers: GPT-4o mini is exceptional value for classification, routing, and simple extraction tasks. For a business processing 500 emails per month with AI classification, the API cost is approximately $0.05–$0.25 — less than a cent per email. GPT-4o is appropriate when you need the highest quality reasoning or generation, but at 30x the cost of GPT-4o mini, it should be reserved for tasks where the quality improvement genuinely matters.
The practical model selection rule
Use GPT-4o mini (or Claude 3 Haiku) for: email classification, lead scoring, content categorisation, simple data extraction, FAQ routing, sentiment analysis, and any task where "pick the right category from a list" describes the core requirement. Use GPT-4o (or Claude 3.5 Sonnet) for: generating actual reply text, processing complex multi-page documents, multi-step reasoning, nuanced tone-matching, and tasks where you consistently find GPT-4o mini's outputs require significant editing. Test both on your specific use case — sometimes GPT-4o mini's outputs are indistinguishable from GPT-4o's for your particular task, and the 30x cost difference becomes a straightforward decision.
Budget tiers: what you can build at each level
- Make.com free (1,000 ops/month) for basic workflow orchestration without AI processing
- ChatGPT free for manual AI-assisted tasks (not automated — requires you to be present)
- Zapier free (5 Zaps, 100 tasks) for very low-volume simple automations
- n8n self-hosted on a $0 local server (your own computer, not publicly accessible)
- OpenAI API: small initial credit for new accounts (~$5 worth)
- Honest assessment: Adequate for learning and experimentation. Not sufficient for production AI automation that processes meaningful volume without any cost.
- Make.com Core ($9/month, 10,000 ops) — unlimited scenarios, 1-min intervals, handles 2,000–2,500 items/month in 4-op workflows
- OpenAI API: $5–$15/month for typical individual use (500–2,000 AI-processed items using GPT-4o mini)
- Builds and runs: email classification and drafting, meeting summarisation, lead scoring, weekly reports, content repurposing, daily briefings
- Alternative: n8n self-hosted ($5/month VPS) + OpenAI API ($5–$15) = $10–$20/month with unlimited operations
- What this looks like in practice: A freelancer or solopreneur running 5–6 active automations processing 500–1,500 items per month across email, leads, and content. Total cost: $14–$24/month.
- Make.com Core ($9/month) or Pro ($16/month for higher ops and team features)
- OpenAI API: $20–$50/month for 2,000–10,000 AI-processed items (mix of GPT-4o mini and GPT-4o)
- Additional specialist tools: Otter.ai Pro ($17/month for team meeting transcription), or Clearbit/Apollo for lead enrichment ($50+/month)
- Builds and runs: all individual automations plus team-shared email workflows, automated customer support (first-response drafting), batch invoice processing, multi-channel content distribution
- What this looks like in practice: A 3-5 person professional services firm running shared email automation, automated client briefings, and content pipelines. Total cost: $40–$70/month depending on enrichment API needs.
- n8n Cloud ($20/month) or Make.com Team/Pro ($29+/month)
- OpenAI API: $50–$150/month for 10,000–50,000 AI-processed items
- Vector database (Pinecone starter $70/month, or Supabase $25/month) for RAG pipelines
- Enrichment and specialist APIs: $50–$200/month depending on volume
- Builds and runs: full customer support automation with RAG knowledge base, multi-stage lead qualification with enrichment, automated reporting across multiple data sources, AI-powered CRM workflows
- What this looks like in practice: A mid-size SaaS company automating customer success workflows, content operations for 20+ pieces per week, and sales qualification for 500+ monthly inbound leads.
Real cost examples: what specific automations actually cost per month
Here are specific monthly cost calculations for the most common AI automation use cases, based on typical volumes for each scenario.
Monthly cost calculator for common AI automations
| Automation | Volume/month | Model | API cost | Make.com ops | Total cost |
|---|---|---|---|---|---|
| Email classification only | 500 emails | GPT-4o mini | ~$0.25 | ~1,000 | ~$9.25 (Core) |
| Email classification + draft | 500 emails | GPT-4o | ~$7.50 | ~1,500 | ~$16.50 |
| Lead scoring from form | 200 leads | GPT-4o mini | ~$0.20 | ~1,000 | ~$9.20 |
| Meeting summaries | 20 meetings | GPT-4o | ~$6.00 | ~100 | ~$6–15 (Otter.ai) |
| Blog to social (5 posts/week) | 20 posts | GPT-4o | ~$4.00 | ~140 | ~$13 |
| Invoice extraction | 200 invoices | GPT-4o | ~$15.00 | ~1,000 | ~$24 |
| Weekly performance report | 4 reports | GPT-4o | ~$1.00 | ~60 | ~$10 |
| Customer FAQ responder (RAG) | 500 queries | GPT-4o mini | ~$1.50 | ~2,500 | ~$11 + Pinecone |
All costs assume Make.com Core plan ($9/month) is already in use. API costs are estimates based on typical prompt and output lengths — actual costs vary with prompt design. Multiple automations share the same Make.com plan cost.
The most expensive automations per item
Document processing (invoices, contracts, reports) is the most expensive AI automation per item because documents are longer — more input tokens — and require more complex extraction — more output tokens. A 2-page invoice processed with GPT-4o Vision might consume 2,000+ input tokens versus 300 tokens for a typical email classification. At GPT-4o's $5/million input tokens, this is $0.01 per invoice — still very affordable in absolute terms, but significantly higher per item than simple text classification.
For document-heavy use cases, consider: using GPT-4o mini for document processing where its extraction quality is adequate (test on 20 documents); pre-processing documents to extract only the relevant sections before sending to the AI (a Make.com HTTP module calling a PDF text extractor reduces token count); or using a specialist document extraction service (Mindee, Reducto) that is optimised for specific document types and often cheaper per extraction than general-purpose LLMs.
Cost optimisation: getting more automation for your budget
Model tiering by task complexity
The single highest-impact cost optimisation for most people: audit which tasks use GPT-4o and migrate any that do not genuinely need its reasoning capability to GPT-4o mini. For a typical portfolio of automations, this often reduces API costs by 50–70% with no measurable quality impact on the tasks where GPT-4o mini's outputs are adequate. Test both models on 20 real examples for each automation — the comparison typically reveals which tasks benefit from GPT-4o and which are indistinguishable.
Batch processing instead of item-by-item
Some automation use cases — weekly reports, content adaptation batches, lead scoring for accumulated form submissions — do not require item-by-item real-time processing. Batching these to run once daily or weekly instead of individually reduces operations consumption (fewer scenario runs) and can also reduce API costs if you can consolidate multiple items into a single API call with structured output.
Input text preprocessing
Before passing content to the AI, strip it down to only what the AI needs. For email classification, the subject line and first 200 characters of the body is typically sufficient — you do not need the full 1,000-word email body. For document extraction, extract specific sections (header, line items, totals) rather than passing the full document. Preprocessing in Make.com's text manipulation modules or a lightweight JavaScript snippet in n8n is free — the API token savings it generates can be substantial.
Response caching for repeated inputs
If your automation processes inputs that repeat frequently — the same FAQ question phrased multiple ways, lead scoring for companies you have scored before — caching previous AI responses and returning the cached result for similar inputs can significantly reduce API costs. This is a more advanced optimisation that requires a simple lookup table (a Google Sheet or a database) and a similarity check before the API call. Worth implementing for high-volume automations with predictable input patterns.
The $9/month challenge: what you can build on Make.com Core + $5 API
With $14/month total, you can run: email classification and labelling for up to 500 emails (using GPT-4o mini at $0.25 total API cost), weekly report generation for 4 reports ($0.40 API cost), content repurposing for 20 blog posts to social formats ($0.80 API cost), lead scoring for 200 inbound leads ($0.20 API cost), and a daily morning briefing from 5 RSS sources ($0.90 API cost). Total API: approximately $2.55. Make.com Core: $9. Total: $11.55/month for automations that a conservative estimate would value at 8–12 hours of saved time per week.
Frequently asked questions about AI automation costs
The minimum for a functional production AI automation that processes real inputs without you being present: Make.com Core ($9/month) + OpenAI API (minimum spend is essentially $0 for very low volume — $5 gets you through several months of light experimentation). For the first month of testing and building: $9 for Make.com and perhaps $1–$2 in API costs. For ongoing production at modest volume: $12–$15/month total. The Make.com Core plan is the essential upgrade from free because it gives you 10,000 operations and 1-minute polling intervals — both critical for real automation work.
Partially. Make.com's free tier (1,000 ops/month, multi-step workflows) is genuinely useful for non-AI automations and for low-volume AI automations where the API cost is your only expense. Zapier's free tier is too limited for meaningful AI automation. n8n self-hosted is free for the software but requires a server (~$5/month). The honest truth: for AI processing in automated workflows (not manual ChatGPT use), you need the OpenAI API — there is no production-capable free alternative for programmatic access to high-quality AI models, though the costs are very low.
Three safeguards that together make unexpected bills essentially impossible: (1) Set a hard monthly spending limit in OpenAI's platform settings (Settings → Limits → Monthly budget) — when reached, API calls simply stop; (2) Add a Make.com filter step that skips items with unusually long inputs (a 50,000-word document accidentally routed to your automation would cost significantly more than a typical email); (3) Monitor your usage in OpenAI's dashboard weekly for the first two months of a new automation. After you understand your actual usage patterns, you can relax the monitoring frequency.
For most people, no. Make.com Core ($9/month) gives you 10,000 operations, 1-minute polling, and unlimited multi-step scenarios — everything Zapier's Starter plan offers at $29.99/month, at a third of the price. The exception: if you specifically need an integration that only Zapier has (given its 6,000+ app library vs Make.com's ~1,500), the Zapier Starter plan may be necessary for that specific workflow. For general AI automation work, Make.com Core is significantly better value.
The crossover typically occurs at very high operation volumes where per-operation platform costs exceed the infrastructure cost of a custom solution. For Make.com Core: at 10,000 operations/month, the per-operation cost is $0.0009. A lightweight Python application running on a $10/month server could handle the same volume at a small fraction of that cost at the platform level — but only after accounting for development time, maintenance time, and the opportunity cost of not building other things. For volumes under 100,000 items per month, no-code platforms are almost always the better economic choice when development and maintenance costs are included.
Start building within your budget today
The complete AI automation guide covers every tool, workflow, and cost optimisation strategy — from free experimentation to enterprise-scale deployments.
Read the Complete AI Automation Guide →ThinkForAI Editorial Team
All pricing in this article verified as of November 2024. API pricing changes frequently — check platform pricing pages before making budget commitments.
