Introducing an internal AI chatbot with predictable costs? The AI platform is a dedicated Open WebUI instance per organisation with models operated in Germany by German providers, SSO and a Section 203 supplementary agreement. Packages from €149.90 excl. VAT per month (Start, up to 10 users), provisioned within a few working days. View the AI platform
An internal AI chatbot, also called a company chatbot or AI workplace, is a chat for employees with models, files and optionally company knowledge. It is not the customer chatbot on a website, for which vendors state their own pricing models, for example moinAI from about €725 per month (vendor page, updated 27 May 2026). This guide calculates the internal chatbot for 10, 25 and 100 users. Every price has a source and the retrieval date 28 October 2026, every assumption is in a table. A price comparison is not a quality comparison: models, functions and data paths of the offers differ. As of October 2026.
Table of contents
- Seven cost types
- Assumptions of the calculations
- Licence per user: ChatGPT for business and Microsoft 365 Copilot
- Model tokens: the cost of an API
- Platform with models in Germany
- Own hardware or rental
- Total comparison over 36 months
- Break-even as a calculation method
- Decision table: three architectures
- Costs that appear in no price list
- Status and method
Seven cost types
| Cost type | Calculation basis | Example in this guide |
|---|---|---|
| Licence per user | users × price × months | ChatGPT Business, Microsoft 365 Copilot |
| Model tokens | requests × tokens × price per million | model API |
| Model operation | fixed monthly amount, independent of user count | AI Cube, Managed GPU Server |
| Platform | interface, sign-in, backup, updates | included in the AI platform package |
| Operations | hours × rate or service price | AI Cube Care, operation by the provider |
| Setup | once: knowledge sources, permissions, assistants | onboarding package |
| Training and rollout | once and ongoing: rules, data classes, works council | two hours of training in the onboarding package |
Which cost types become visible depends on the architecture. A SaaS chatbot bundles tokens, model operation, platform and operations into the licence price, an own setup separates them. With the branding left unchanged, Open WebUI itself causes no licence costs (Open WebUI License 2026).
Assumptions of the calculations
| Assumption | Value |
|---|---|
| Period | 36 months |
| Amounts | net or as shown on the pricing page, retrieved on 28 October 2026. Microsoft and WZ-IT state prices plus VAT, the OpenAI pricing page and the AI Model Hub section at IONOS make no statement on VAT. US dollar prices are not converted and not summed |
| Users | named users: 10, 25 and 100, matching the package sizes of the AI platform |
| Request | 5,000 tokens, of which 4,000 input and 1,000 output (ratio 4 to 1, as in What does a local AI server cost?) |
| Usage profiles | occasional 100, regular 400, intensive 1,000 requests per user and month, that is 0.5, 2 and 5 million tokens |
| Not included | electricity, internal working time, integrations, data protection and works council procedures |
Licence per user: ChatGPT for business and Microsoft 365 Copilot
| Offer | Price per user and month | 10 users | 25 users | 100 users |
|---|---|---|---|---|
| ChatGPT Business, standard seat | €21 | €210 | €525 | €2,100 |
| Microsoft 365 Copilot Business | €18.20 | €182 | €455 | €1,820 |
OpenAI shows €21 per standard seat and month for Germany, and in US dollars 20 billed annually and 25 billed monthly. The page does not say which billing type the euro amount applies to. ChatGPT Business is for teams of 2 to 200 employees. ChatGPT Enterprise has no stated price and lists SCIM, SLAs and data residency in ten regions, among other things.
Microsoft states €18.20 per user and month for Copilot Business billed annually plus VAT, with a promotional price of €15.60 on 28 October 2026 (first year only, offer until 31 December 2026), as an addition to an existing Microsoft 365 plan. For large enterprises there is Microsoft 365 Copilot at €26.00. Copilot Chat is available at no extra cost to users with an eligible Microsoft 365 subscription; the page lists the AI chat based on business data only in the paid plan. Licence costs grow linearly with the number of users.
Model tokens: the cost of an API
| Model (provider) | Input / output per million tokens | blended (4 to 1) | per user at 2M tokens | 10 users | 25 users | 100 users |
|---|---|---|---|---|---|---|
| gpt-oss-120b (IONOS) | €0.15 / €0.65 | €0.25 | €0.50 | €5 | €12.50 | €50 |
| Qwen3.5-397B-A17B (IONOS) | €0.60 / €3.60 | €1.20 | €2.40 | €24 | €60 | €240 |
| GPT-6.1 Terra (OpenAI API) | $1.00 / $5.00 | $1.80 | $3.60 | $36 | $90 | $360 |
| GPT-6.1 Sol (OpenAI API) | $2.00 / $10.00 | $3.60 | $7.20 | $72 | $180 | $720 |
Blended price = (4 × input + output) ÷ 5. At 0.5 million tokens per user the amount drops to a quarter, at 5 million it rises to 2.5 times. The OpenAI prices apply to standard processing, and OpenAI charges a 10 % uplift for regional processing (data residency). Cached inputs and batch prices are not included.
With these profiles the token item stays small compared with seat licences. It grows with usage and model choice: Qwen3.5-397B costs 4.8 times as much as gpt-oss-120b. A model API alone is not yet a chatbot; interface, sign-in, operations and knowledge sources come on top.
Platform with models in Germany
| AI platform package | Included users | Price per month | 36 months | with onboarding package | per user and month | Tokens per month (fair use) |
|---|---|---|---|---|---|---|
| Start | 10 | €149.90 | €5,396.40 | €6,386.40 | €14.99 | 75M |
| Team | 25 | €399.90 | €14,396.40 | €15,386.40 | €16.00 | 300M |
| Organisation | 100 | €1,199.90 | €43,196.40 | €44,186.40 | €12.00 | 1bn |
The monthly price includes SSO, a .de domain (or your own domain via CNAME), one hour of introduction, DPA and Section 203 supplementary agreement, updates, daily backup and monitoring. Instance and models run in Germany with German providers. The onboarding package costs €990 once (optional), sets up the first knowledge base together with you and includes two hours of training. Each additional user costs €14.90 per month. The Business package for 50 users costs €699.90.
The “intensive” profile also stays below the token allowance: 50 of 75, 125 of 300 and 500 of 1,000 million tokens. The platform offers a fixed model list, which is shown on the product page.
Own hardware or rental
| Offer | Costs | 36 months | Monthly average | per user at 10 | at 25 | at 100 |
|---|---|---|---|---|---|---|
| AI Cube with Care | from €7,990 once (1 TB version), Care €349.90 per month (12-month minimum term) | €20,586.40 | €571.84 | €57.18 | €22.87 | €5.72 |
| Managed GPU Server 24 | €699 per month, €499 once | €25,663.00 | €712.86 | €71.29 | €28.51 | €7.13 |
| Managed GPU Server 96 | €1,799 per month, €999 once | €65,763.00 | €1,826.75 | €182.68 | €73.07 | €18.27 |
These amounts are fixed costs and depend on neither users nor tokens. Whether a system carries 10, 25 or 100 users is decided by concurrently active users, model and context length, not by the number of accounts, see How many users can a local AI server support?. The table therefore only divides costs and says nothing about capacity. Electricity at your own site comes on top for the AI Cube (about 150 W under load according to the product page). The derivation of the hardware prices is in What does a local AI server cost?.
Total comparison over 36 months
| Variant (net, 36 months) | 10 users | 25 users | 100 users |
|---|---|---|---|
| ChatGPT Business, standard seat | €7,560 | €18,900 | €75,600 |
| Microsoft 365 Copilot Business | €6,552 | €16,380 | €65,520 |
| AI platform, package | €5,396.40 | €14,396.40 | €43,196.40 |
| AI platform with onboarding package | €6,386.40 | €15,386.40 | €44,186.40 |
| AI Cube with Care | €20,586.40 | €20,586.40 | €20,586.40 |
| Managed GPU Server 24 | €25,663.00 | €25,663.00 | €25,663.00 |
| Managed GPU Server 96 | €65,763.00 | €65,763.00 | €65,763.00 |
Three findings from the table:
- Seat licences grow linearly. The platform packages rise in steps, and the price per user, when fully used, lies between €12.00 and €16.00.
- Hardware is independent of users. At 10 users the fixed amount per user is highest, at 100 it is lowest, provided the load is carried. Running on own hardware removes token costs.
- The table shows prices, not performance. ChatGPT offers current OpenAI models and vendor functions, the platform a fixed model list in Germany, hardware free model choice within the memory. Integration with knowledge sources comes on top in every variant.
Break-even as a calculation method
Flat price ÷ seat price = number of users from which the flat price is below the seat costs. With the prices above: Start package €149.90 ÷ €21 = 7.1, so from 8 users. Against Copilot Business €149.90 ÷ €18.20 = 8.2, so from 9. For hardware the averaged monthly amount is used: AI Cube €571.84 ÷ €21 = 27.2, so from 28 users.
The result is a calculation value, not a recommendation. It assumes equivalent functions, equivalent model quality and sufficient capacity, and it changes with every price. For self-operation the same calculation runs backwards: additional cost compared with the service price ÷ internal hourly rate = hours per month that self-operation may cost before it becomes more expensive.
Decision table: three architectures
| Criterion | SaaS workplace | Platform with models in Germany | Own hardware |
|---|---|---|---|
| Cost form | per user and month | package by user count, additional users individually | fixed amount, independent of users |
| Data path | provider cloud, data residency depends on the plan | instance and models in Germany with German providers | AI Cube on your own network, GPU server dedicated in Germany |
| Model choice | set by the provider | fixed model list | free within the hardware |
| Operations | at the provider | at WZ-IT | own team or WZ-IT |
| Start | account and sign-in | a few working days, 7-day test instance | AI Cube: delivery in ten working days according to the product page |
| Fits better when | few users, cloud data is permitted | mid-sized teams, data in Germany, no own hardware | stable load, data stays on your own network, offline operation |
| Limit | functions and region depend on the plan, costs linear | models on the list, token allowance with fair use | capacity and sizing |
Costs that appear in no price list
These items arise in every architecture. They are time, not a list price, and belong in your own calculation:
- Data protection review: purpose, legal basis, processing agreement, see GDPR-compliant AI: assessment criteria.
- Co-determination: coordination with the works council, see AI assistants and the works council.
- Usage rules and AI literacy: measures under Art. 4, see EU AI Act for companies.
- Knowledge sources with permissions: RAG with permissions.
Status and method
- Status: 28 October 2026. Retrieved on 28 October 2026: OpenAI Business, Microsoft Copilot Business and Enterprise, IONOS AI Model Hub, OpenAI API, moinAI. WZ-IT prices from the price list, net, as of October 2026.
- Calculation: licence = users × price × 36. Hardware = one-off amount + 36 × monthly amount, averaged over 36. Tokens = requests × 5,000 × blended price. Rounded to two decimals.
- Deviations: the OpenAI page does not explain the euro amount, the Microsoft page showed a promotional price. Both figures change without notice.
- Not included: volume discounts, enterprise contracts, time savings and benefit. The guide compares costs, not impact.
Tables and figures may be reused with attribution and a link to this article.
Our approach at WZ-IT
It starts with the usage profile: named users, concurrently active users, data classes and knowledge sources. The architecture follows from that. For an entry without hardware there is the AI platform: packages from €149.90 excl. VAT per month for up to 10 users, a test instance for 7 days and provisioning within a few working days.
If model and knowledge should stay on your own network, WZ-IT delivers the AI Cube from €7,990 excl. VAT once, with operations covered by AI Cube Care from €349.90 excl. VAT per month. A dedicated system in a data centre in Germany is the Managed GPU Server from €699 excl. VAT per month. Which tier fits the load, we check before the quote.
Further guides
- Private ChatGPT for business - four operating models compared
- What does a local AI server cost? - break-even per token
- How many users can a local AI server support? - load test instead of account count
- Local AI or ChatGPT Business? - comparison of cost, control and use
- European LLM APIs compared - providers, locations and models
- Cloud AI vs. self-hosted - the basic trade-off
Want your user count and data classes assessed? We calculate the variants with your usage profile. AI platform · Book a call
Further reading: Web search in Open WebUI: providers compared
