EnroutiaEU
ComparatorPricingDocsSign in

Write one prompt. Read what 8 models answer — GPT and Claude included.

Paste the prompt you would give ChatGPT. Enroutia runs it against 5 open models we serve from European clouds and 5 closed ones we run only to compare, and shows you every answer in full, side by side — with what 1,000 runs would cost and how long each took. You judge the answers; we only put them next to each other.

No sign-up. Five comparisons a day, twenty with a confirmed email. Your prompt is never stored — anywhere.

Compare your prompt

Your prompt against 8 models, right now

The box opens with a recorded demo so the first click shows a full comparison. Edit a single character and it runs live against every provider.

Or try
358/2000Demo prompt: edit it and the comparison runs live.We do not store your prompt. Anywhere.

How to read it

  1. 01

    Write the prompt

    Your real one, or a ready-made task from the chips. The recorded demo replays instantly; anything you type runs fresh against each provider, streaming as the models answer.

  2. 02

    Read the answers

    Every card carries the model's name and its answer in full — nothing summarised, nothing scored for you. The closed models are marked: we run them so you can compare, we do not sell them.

  3. 03

    Look at the two numbers

    What 1,000 runs of this exact prompt would cost at each model's price, and how long each took to finish. The cheapest of ours is marked; whether its answer is good enough is a call only you can make.

Ready-made tasks

Six prompts most automations run. Each page opens with its comparison already running.

The same thing, over your whole n8n

One prompt is a demo. The free audit replays up to 200 real executions per task from your own history against cheaper models, and tells you task by task where the output is identical and the price is not.

  1. 01

    Import the collector in your n8n

    One importable workflow. It samples up to 200 stored executions per task, and pseudonymisation runs inside your instance — the mapping table never leaves it.

    What the collector keeps, and where
  2. 02

    Get your audit key

    Leave an address. Audits are approved manually while we run the first cohort. You'll hear from a person, once.

  3. 03

    See your number in ~15 minutes

    The replay runs your samples through the whole gateway — resolution, failover, retries — and a verdict engine scores every candidate against your production output. The report lands in your inbox.

    Open the full report

European privacy, by architecture

We cannot read your production prompts: nothing is stored — a request lives in memory for the duration of the call and is gone.

Pseudonymisation happens inside your n8n instance before anything is sent: emails, phone numbers, IBANs, IDs, card numbers and any field you flag become stable tokens, and the mapping table never leaves your machine.

Audit samples are the second of two declared exceptions to that rule — what we keep, why, for how long and how it is deleted is published, not implied.

How our privacy works

Frequently asked questions

Which models are these?
Five open-weight models — Mistral, Llama, Qwen, DeepSeek and gpt-oss — served from European clouds and available through Enroutia at the prices on the pricing page. Plus three closed ones, GPT and two Claude, which we run for the comparison only: they are marked on their cards, and we do not sell them.
Are the answers real?
Yes. Every run goes live to each provider and streams back as it is written. The one exception is the demo prompt the box opens with: it replays a recording made on the date shown, so the first click costs nobody anything. Edit one character and it runs live.
Why is there a daily limit?
Because each comparison buys eight models' worth of tokens, and here we pay for them. Five a day per visitor, twenty with a confirmed email, and an account compares the whole catalogue at your own prices.
Do you store my prompt?
No. It lives in memory for the duration of the call and is gone. What we record is that a comparison happened — a daily-rotated hash of the address, which models ran and how long they took — never a word of what you typed or what came back.
What does “per 1,000 runs” mean?
The price of this exact prompt and this exact answer, times a thousand — our models at our price, the closed ones at their providers' published list prices. It is the number a workflow costs, which is the one you can act on; a price per token is not.
What is the audit?
The same comparison, over your real executions instead of one prompt. A collector you import into your n8n samples up to 200 stored executions per task, pseudonymised inside your instance, and the replay scores every candidate against your production output. It is free; we charge only when you switch, as a share of verified savings.
What if there's nothing to save?
Then the report says so, and it costs you nothing. Some tasks are already on the cheapest model that can do them, and for those the honest verdict is “don't switch”. The first audit we ran was our own, and that is exactly what it returned. An audit that can only ever find savings is a brochure, not an audit.