Stop overpaying for your prompts.
Paste the prompt you run every day. Open models answer it alongside GPT-5.6 and Claude — every name hidden. Pick the answer you would actually ship, and we will show you what each one costs. It is rarely the expensive one.
Cheapest run on the panel takes all the points; the rest score in proportion.
Time to finish the answer, measured on this run, not on a leaderboard.
The answer you picked. The one part of the score a machine has no business guessing.
Three steps, and one of them is copying a key.
There is no SDK to install and no new API to learn. If your automation already talks to OpenAI, it already talks to us.
- 01
Take your key
One key per project or per client, each with its own spending cap and alert. Twenty of them if you want.
- 02
Change one line
Point the base URL at Enroutia. n8n, Make, WordPress, the OpenAI SDK — anything that already speaks to OpenAI works unchanged.
- 03
Pay per message
Prepaid credit in euros, priced per 1,000 messages. No subscription, no minimum, no invoice you did not see coming.
client = OpenAI(
- base_url="https://api.openai.com/v1",
+ base_url="https://api.enroutia.com/v1",
api_key=os.environ["ENROUTIA_KEY"],
)Priced per thousand messages, not per million tokens.
A message is 600 tokens — about what a normal email and its reply take up. Prepaid credit, no subscription, and the meter stops when the balance does.
We pick the best open model for the price. Today that is gpt-oss-120b, and we move it as better models appear.
Fast and cheap. Ideal for classification, data extraction and short replies.
Balanced quality and price. Our default choice.
Thinks before it answers: slow to start, and the thinking is billed even though you never see it. The best here for code and multi-step problems.
Natural writing and reliable instruction following.
The most capable model here for complex, multilingual work.
GPT-5.6 and Claude answer your prompt on the panel above and report their own token usage, so the comparison is what that run actually cost, not an estimate. Their per-token rates are the providers’ published list prices, reviewed on 2026-08-18 and converted to euros. We do not resell either one.
Four claims, all of them checkable.
European by architecture, not by policy
Inference runs in Paris on Scaleway. Your data does not leave the EU because there is nowhere else for it to go.
We cannot read your prompts
They live in memory for the length of the request and disappear. No content logs, no training set, nothing on disk to subpoena.
No latency of our own
Routing adds under 30 ms at the 95th percentile, measured continuously and published. The proxy sits in the same region as the models.
Nothing to learn
It is the OpenAI API. Same request shape, same streaming, same error codes. If your code works today it works here.
We are not open yet. On purpose.
The platform is running today for our own workloads while we settle pricing against real margins. Leave an address and we will write once — when there is a key to hand you, not before.
One email. No newsletter, no drip sequence, no sharing your address with anybody.