Test mode: payments run through the Stripe sandbox. Nothing is charged, reservations do not count in the counter.

Skip to content
llmcrowd
DE Deutsche VersionReserve

No more token bills.

The AI workspace for companies from 10 people: chat, parallel agents and your data on our own servers in Germany and Switzerland. One fixed price per seat.

Reserve seatsDo the maths

Planned start of server 1 · 1 April 2027

–days
–hrs
–min
–sec

Server CH-1: 0/400 seats reserved · Switzerland

Planned, not a guaranteed delivery date: server 1 is ordered once the first 400 seats are reserved.

Reserve for 99 per seat, from 10 seats, fully refundable. Then 200 per seat and month, CHF or EUR, excluding VAT.

How it works

Over 1,000 US dollars of API value. You pay 200.

That is what a power user's work with agents costs per seat and month when bought through the APIs of frontier models at list prices. Check it with your own usage.

Usage

8M in, 0.8M out per day: long documents (200k to 1M context), several parallel agents, repeated runs

Model

4 / 20 US dollars per million tokens[2]

API value of this usage, per seat and month

1,008US$

llmcrowd, per seat and month

200 fixed

CHF in Switzerland, EUR in the EU, annual payment, excluding VAT.

Your value: at least 4× what you pay.

For 25 seats per year: US$302,400 through the API against 60,000 CHF/EUR with llmcrowd.

Basis: official list prices (input/output) without volume discounts, 21 working days a month. Caching and batch prices can lower API costs. Factor stated conservatively: dollar amount reduced by 20% for exchange rates, rounded down. llmcrowd runs Kimi K3.1 Pro, not the models named above. In the vendor's tests, K3.1 Pro is ahead of Claude Opus 5.5 in 16 of 37 benchmarks. The figure shows what this usage would cost with the named models at list prices.[10]

Your data does not leave your country.

Every country gets its own servers and its own tranches: your prompts are processed on servers in your country. We start in Switzerland and Germany; more countries follow based on waitlist entries.

Switzerland · Reservation open

400seats free on server CH-1

Server CH-1: 0/400 seats reserved

Companies: 0

400 seats left in tranche CH-1

  1. Server CH-1 · 400
  2. Server CH-2 · 400
  3. Server CH-3 · 400
  4. Server CH-4 · 400
  5. Server CH-5 · 400
  6. Server CH-6 · 400
  7. Server CH-7 · 400
  8. Server CH-8 · 200

Expansion in tranches: one server per 400 seats, up to 3,000 seats. Only paid reservations and letters of intent confirmed in writing are counted, no test payments.

Planned start of server 1 · 1 April 2027

–days
–hrs
–min
–sec

Planned, not a guaranteed delivery date: server 1 is ordered once the first 400 seats are reserved.

Reserve seats

The whole application, not just a model.

You get a finished workspace, not an API key. Everything runs in your environment, everything is logged.

A pool instead of hard limits.

For every package we build a dedicated server environment in Germany or Switzerland. The companies in that environment share its capacity. When your team works a lot and others little, you use more. The servers run around the clock.

No lock screen
No quota that runs out after five hours, no extra charge per token.
Fair use, explained openly
When the pool is full, it shares compute fairly: every seat gets at least its share, chat stays fast, background agents wait briefly. Nobody can drain the pool alone.
Why from 10 seats
We build a dedicated environment for every customer. That is why llmcrowd starts at 10 seats, not as a single seat.
Your team other companies freeExplanatory simulation, not live data.

Cheap subscriptions come with a catch.

Around 100 per seat and month buys quotas that pause until a reset, extra charges by usage, and US groups processing your prompts. That is what the vendors' own pricing pages say.

  • llmcrowd

    Price
    CHF 200 or EUR 200 per seat and month, fixed (annual payment)
    Usage limit
    No quota. Shared capacity pool with fair use
    When the limit is reached
    No hard stop. When the pool is full, background agents wait briefly, chat stays fast
    Who processes your prompts
    Our servers in Germany or Switzerland, no US group in the chain

  • Claude Max

    Price
    100 US dollars per month, individual[1]
    Usage limit
    Limits every 5 hours and every week[1]
    When the limit is reached
    Wait for the 5-hour window or the week to reset[1]
    Who processes your prompts
    Anthropic, US company, CLOUD Act[9]

  • Claude Team Premium

    Price
    100 US dollars per seat and month paid annually, 125 monthly[1]
    Usage limit
    Limits per seat, every 5 hours and every week[1]
    When the limit is reached
    Wait for the reset[1]
    Who processes your prompts
    Anthropic, US company, CLOUD Act[9]

  • Claude Enterprise

    Price
    20 US dollars per seat plus all usage at API rates, from 20 seats[1]
    Usage limit
    Spend limits set by the admin[1]
    When the limit is reached
    Every further use is billed at token prices[1],[2]
    Who processes your prompts
    Anthropic, US company, CLOUD Act[9]

  • ChatGPT Business

    Price
    Standard 20 US dollars, Premium 100 US dollars per seat and month paid annually (monthly 25 or 125)[4]
    Usage limit
    Quotas per seat («Limits apply»), Premium with a higher quota[4],[3]
    When the limit is reached
    Buy credits[3]
    Who processes your prompts
    OpenAI, US company, CLOUD Act[9]

  • Microsoft 365 Copilot Business

    Price
    21 US dollars per seat and month (promotion 18), on top of the Microsoft 365 licence[5]
    Usage limit
    Selected AI features billed by usage[5]
    When the limit is reached
    Extra charges by usage[5]
    Who processes your prompts
    Microsoft, US company, CLOUD Act[9]

  • Google Gemini

    Price
    Included in Workspace plans, scope per plan[6]
    Usage limit
    Limited in some plans[6]
    When the limit is reached
    A larger plan is needed[6]
    Who processes your prompts
    Google, US company, CLOUD Act[9]

With the offers above, your prompts are processed by US groups. With llmcrowd they run on our servers in your country, with your own database, an audit log and a model you can switch by contract.

List prices and limits as on the linked vendor pages, as of 10 November 2026, without guarantee. Prices excluding VAT. No statement about model quality.

One price. From 10 seats.

Every feature in every seat: chat with 1 million tokens of context, parallel agents, connectors, your own database, audit log and DPA. 12-month term, prices excluding VAT.

Recommended · saves €15,000 with 25 seats

Annually in advance

200

CHF or EUR per seat and month

2,400 per seat and year, in advance

Reserve, annual

Monthly instalments

250

CHF or EUR per seat and month

12 months fixed, not cancellable monthly (25% surcharge)

Reserve, monthly

Enterprise

Individual

From 400 seats: your own capacity pool and server environment.

Get in touch

The first payment only becomes due once the tranche you reserve in is full (400 seats = one server). The 99 per seat is credited.

Where your money goes

Calculator

CHF / EUR

Reservation today

€2,475

99 per seat, fully refundable

Later annually
€60,000
per year
Or monthly
€6,250
per month, 12 months

Saving with annual payment: €15,000 per year

Reserve 25 seats

Data protection your DPO can check.

Nobody can seriously promise complete legal certainty. We design operations to comply with the GDPR and show how.

DE · CH
Servers in the country of use
Art. 28
Data processing under the GDPR
0
US groups in the chain

Processing agreement under Art. 28 GDPR

DPA with technical and organisational measures and the list of all sub-processors.

Data stays in DE/CH

Processing and storage in data centres in Germany or Switzerland, no US group in the chain.

Deletion concept

Fixed periods per data type, deletion on request, export of your data at any time.

Audit log

Every question, file access and agent action is logged and visible to you.

No training on your data

The model runs unchanged on our servers. Your prompts and files serve only your request.

Tenant separation

Every company has its own database; agents only get the permissions you grant.

The contracts (DPA, terms, reservation terms) are reviewed by a lawyer before launch.

Where the model comes from. Said openly.

llmcrowd runs Kimi K3.1 Pro, an open-weight model by Moonshot AI from China, with 1 million tokens of context. Open-weight means the model weights run on our own servers. There is no connection to Moonshot, no data goes to China.[13]

In the vendor's tests, K3.1 Pro beats Claude Opus 5.5 in 16 of 37 benchmarks and trails in some tasks. That is why we show you the model in a demo on your own tasks before anything is due.[10],[11]

The GPUs come from NVIDIA in the US. If you reject a model from China, we switch to another open-weight model, guaranteed by contract.

16 of 37 benchmarks ahead of Claude Opus 5.5, per the vendor. It trails in the others.

The founders' tranche: what happens next.

We start with one server and expand in tranches. We only buy hardware once a tranche is full.

  1. 01

    Reserve

    €99 per seat, from 10 seats. Your payment method is saved for later.

  2. 02

    Tranche fills

    400 seats make one server. The counter above shows how far server 1 is.

  3. 03

    Due and ordered

    Once the tranche is full, the first payment becomes due (less €99 per seat) and we order the server.

  4. 04

    Go-live

    Set-up, testing, launch. Your 12-month term begins.

If server 1 has not started by 30 June 2027, we refund every reservation automatically and in full.

Questions managing directors ask us.

Get in touch
Is the model as good as ChatGPT or Claude?

We do not promise that. In the vendor's tests, Kimi K3.1 Pro is ahead of Claude Opus 5.5 in 16 of 37 benchmarks and behind in others. Before anything is due, we are happy to show it in a demo on your own tasks.

What exactly does «no hard stops» mean?

There is no quota that runs out and no extra charge. The capacity of a server environment is shared. When it is full, every seat gets at least its fair share and background agents wait briefly. Nobody can seriously promise unlimited compute, and neither do we.

What happens to my €99 per seat?

Once the tranche you reserved in is full, we credit the €99 per seat against your first payment. If the launch does not happen (server 1 not by 30 June 2027), we refund automatically and in full.

When is how much due?

Only once the tranche you reserved in is full: CHF 2,400 per seat with annual payment, or the first instalment of CHF 250 per seat with monthly payment, each less the reservation. We charge your saved payment method and announce it by email beforehand. The term is 12 months either way.

Why not just Azure or AWS in the EU?

Even in EU data centres, Microsoft and Amazon remain US groups to which the US CLOUD Act applies. We run llmcrowd without a US group in the chain.

The model comes from China. Is that a risk?

The model runs entirely on our servers in Germany or Switzerland, with no connection to its maker. If you still want a different model, we switch, guaranteed by contract.

Who is liable when an agent makes a mistake?

Agents act only with the permissions you grant. Changes to production systems need your approval, every action is in the audit log, and there is a kill switch. Liability is governed by the terms.

What happens if a server fails?

From server 2 we run redundantly. Until then the maintenance windows and response times in the contract apply.

From how many employees does it pay off?

From 10 seats. We build a dedicated environment for every customer, so we do not offer single seats.

Yes, count me in.

Reserve seats in the founders' tranche now. €99 per seat, fully refundable.

Reserve seats

Sources

  1. [1] Anthropic, Claude Pricing (Team & Enterprise), retrieved 10 November 2026
  2. [2] Anthropic, API Pricing (Claude Opus 5.5), retrieved 10 November 2026
  3. [3] OpenAI, ChatGPT Pricing (Business & Enterprise), retrieved 10 November 2026
  4. [4] OpenAI Help Center, Managing billing and seats in ChatGPT Business, retrieved 10 November 2026
  5. [5] Microsoft, Microsoft 365 Copilot Pricing, retrieved 10 November 2026
  6. [6] Google, Google Workspace Pricing, retrieved 10 November 2026
  7. [7] OpenAI, API Pricing (GPT-6 Astra), retrieved 10 November 2026
  8. [8] Google, Gemini API Pricing (Gemini 4 Argon), retrieved 10 November 2026
  9. [9] 18 U.S.C. § 2713 (CLOUD Act), Cornell Law School LII, retrieved 10 November 2026
  10. [10] Startup Fortune, Kimi K3.1 Pro: Weights und Benchmarks, retrieved 9 November 2026
  11. [11] MarkTechPost, Kimi K3.1 Pro Open-Weight-Modell, retrieved 9 November 2026
  12. [12] ComputingForGeeks, Kimi K3 lokal betreiben: Hardware und KV-Cache, retrieved 9 November 2026
  13. [13] vLLM, Day-0-Support für Kimi K3.1 Pro, retrieved 9 November 2026