Self-host vs SaaS: AI cost calculator
Short answer
For a 20-person team on ChatGPT Business (Standard seat) (annual billing, $20 per user per month), the subscription costs $4,800 a year. Self-hosting Open WebUI + LiteLLM on a Hetzner Cloud CPX22 server with Claude Sonnet 5.5 API usage at our "typical" placeholder of 2M tokens per user per month comes to about $3,389 a year including 2 hours of upkeep a month at $50/h, so setup time pays back in about 3.4 months. The result swings most with API usage and your own time, so change those inputs first.
Prices checked against vendor pages (dates in the assumptions panel) · server prices from Hetzner Cloud, read 2026-10-22
Assumptions and sources
How the calculator works
- SaaS cost = users × the vendor's published price per user per month (or the plan price for flat plans such as Zapier's task tiers) × 12.
- Self-host cost per month = server + model API + upkeep hours × hourly rate. Year 1 adds the one-off setup hours.
- Server size: we add up the official minimum CPU and RAM of each app in the stack and pick the cheapest Hetzner Cloud plan that fits. If an app publishes no figure, we say so and you choose the server.
- API spend = users × tokens per user × the official per-token list price (Claude Sonnet 5.5 $2/$10, Claude Haiku 5.5 $0.1/$0.5, Claude Opus 5.5 $4/$20 per 1M input/output tokens, source, checked 2026-10-22). The token volumes are placeholders. Enter your own.
- Currency: SaaS prices are mostly USD and hosting is EUR. We convert at the ECB reference rate (1 EUR = 1.1234 USD on 2026-10-21), which you can edit.
Worked examples (default inputs)
| Scenario | SaaS / year | Self-host / year | Break-even |
|---|---|---|---|
| 20 users: ChatGPT Business (Standard seat) vs Open WebUI + LiteLLM on CPX22 | $4,800 | $3,389 (year 1: $3,789) | 3.4 months |
| 50 users: ChatGPT Business (Standard seat) vs Open WebUI + LiteLLM on CPX22 | $12,000 | $6,269 (year 1: $6,669) | 0.8 months |
| 20 users: Microsoft 365 Copilot Business (add-on) vs Open WebUI + LiteLLM on CPX22 | $5,040 | $3,389 (year 1: $3,789) | 2.9 months |
Annual billing; Claude Sonnet 5.5 API at the "Typical (daily use, some documents)" placeholder (2M tokens/user/month); 8 h setup and 2 h/month upkeep at $50/h (placeholders). Server prices exclude VAT.
SaaS prices used
| Product and plan | Price | Source (checked) |
|---|---|---|
| ChatGPT Business (Standard seat) | $20 annual · $25 monthly per user/mo | help.openai.com (2026-10-22) |
| Zapier Team (2,000 tasks/month, up to 25 users) | $69 annual · $103.5 monthly per plan/mo | zapier.com (2026-10-22) |
| GitHub Copilot Business | $19 monthly per user/mo | docs.github.com (2026-10-22) |
| Notion Business (includes Notion AI) | $20 listed per user/mo | www.notion.com (2026-10-22) |
| Otter.ai Business | $24 annual · $30 monthly per user/mo | otter.ai (2026-10-22) |
| Intercom Fin Essential (includes Fin AI Agent) | $29 annual per user/mo | www.intercom.com (2026-10-22) |
| Zendesk AI Suite Team (includes AI agents) | $55 annual · $69 monthly per user/mo | www.zendesk.com (2026-10-22) |
| Microsoft 365 Copilot Business (add-on) | $21 annual · $25.2 monthly per user/mo | www.microsoft.com (2026-10-22) |
| ChatGPT Business (Premium seat) | $100 annual · $125 monthly per user/mo | help.openai.com (2026-10-22) |
| Claude Team (Standard seat) | $20 annual · $25 monthly per user/mo | www.claude.com (2026-10-22) |
| GitHub Copilot Enterprise | $39 monthly per user/mo | docs.github.com (2026-10-22) |
FAQ
Is it cheaper to self-host ChatGPT than to pay for ChatGPT Business?
It depends on team size and API usage. In our worked example (20 users, ChatGPT Business (Standard seat) at $20/user/month billed annually), self-hosting costs about $282 a month against $400. Heavy API usage or a high value on your own time can reverse that. The self-hosted apps themselves are free, but you still pay for the server and for model tokens.
What does the self-hosted cost include?
The server (Hetzner Cloud Regular Performance (shared AMD vCPU), EUR excluding VAT, converted at the ECB reference rate), pay-as-you-go model API spend, monthly upkeep time and a one-off setup time valued at your hourly rate. It excludes VAT, backups, extra storage, GPU servers and SaaS features the open-source apps do not replace.
Where do the SaaS prices come from?
Each price was read from the vendor's own pricing or help page and is shown with its link and the date it was checked. Products whose price we could not verify are left out instead of guessed.
Why are the token volumes called assumptions?
We have not measured how many tokens a typical employee uses, so the light, typical and heavy levels are round placeholders. Replace them with the numbers from your API provider's usage dashboard, or switch to a fixed monthly budget.
How is break-even calculated?
Break-even in months = one-off setup cost ÷ (monthly SaaS cost − monthly self-hosting cost). If self-hosting costs more per month, it never breaks even on cost alone.
Can I share or embed my result?
Yes. Every input is stored in the page URL, so "Copy shareable link" gives a link that reopens the same calculation. The same calculator appears on each stack page under /stacks/.
Related
- Which self-hosted AI app should we use? (4-question wizard)
- Private ChatGPT for a small business (Open WebUI + LiteLLM)
- Private document Q&A over your company's knowledge (Onyx)
- Self-hosted AI workflow automation (a Zapier replacement with n8n)
- Self-hosted meeting transcription (Speaches + Open WebUI)
- Self-hosted coding assistant for your dev team (Tabby)
- Self-hosted customer support AI chatbot (Dify)
- Fully offline private AI chat (Ollama + Open WebUI)
Spotted an outdated price? Tell us and we will re-check the vendor page.