Evanesca vs OpenRouter: one router for every model, or models run in Europe

At a glance

Each OpenRouter cell links to the OpenRouter page it comes from; every one was read on 6 October 2026. Each Evanesca cell links to the Evanesca page that describes it.

Price: OpenRouter usually costs less

Two models are served by both. For Llama 3.3 70B, OpenRouter's headline price is the cheapest of the 10 hosts listed on its page. For GPT-OSS 120B, the row shows the only endpoint OpenRouter offered through its EU domain on 6 October 2026.

OpenRouter publishes its prices in US dollars, before tax; Evanesca's are in euros, margin and 20% VAT included. So the verdict compares both before tax: Evanesca's price divided by 1.2, and the dollar converted at the ECB reference rate of 6 October 2026, 1 EUR = 1.1269 USD (ECB). A gap under 2% names no winner.

On Llama 3.3 70B, OpenRouter costs 93% less on input and 77% less on output. That headline is a floor: the same page lists hosts charging up to $1.04 per million input tokens, and your filters decide which of them you get. On GPT-OSS 120B through the EU domain, OpenRouter costs 37% less on both input and output. The 8% fee of the Business plan, which EU routing requires unless you are on Enterprise, is not in that list price.

Evanesca's price is set per model: what the operator charges, a margin and VAT. It does not follow the cheapest host of the day, and it is exactly what comes off your credit.

What happens to your prompts

OpenRouter's own default is close to Evanesca's. It says it “does not store your prompts or responses, unless you opt in”, either to private logging for your own debugging or to letting OpenRouter use them in exchange for a 1% discount; both are off by default. It keeps request metadata, such as token counts and latency source.

« Any prompt retention on OpenRouter is always opt-in. »

The second layer is the host. “Each AI provider on OpenRouter has its own data handling policies for logging and retention”, and OpenRouter publishes them host by host source. You can stop routing to hosts that may train on prompts, require zero data retention on a single request, or set either rule for the whole account.

On Evanesca, both layers are set in advance. Evanesca stores no prompt and no answer: they are held in memory while the answer is written, your history stays in your browser, and only billing data — tokens and cost — is kept. The operator is named: OVHcloud states it keeps “only the data required for billing purposes” source; Scaleway may keep and read requests that cause errors or look abusive, for up to 14 days source; and Mistral AI keeps its models' inputs and outputs 30 days to monitor abuse source. More on the privacy page.

EU routing answers where, not who

On its Business and Enterprise plans, OpenRouter offers in-region routing: requests sent to eu.openrouter.ai are processed in the EU and, in its words, “never leave it at any point in the request lifecycle”. If no EU endpoint exists for a model, the request fails rather than leaving the region source. That settles where the computation happens.

It does not change which companies handle the request. OpenRouter, Inc. is in New York. On 6 October 2026 its models API listed 466 entries, 70 of them eligible for EU routing source, and the only EU endpoint for GPT-OSS 120B was Amazon Bedrock's eu-west-1 region source, a service of Amazon Web Services, Inc., based in Seattle source. The US CLOUD Act lets US authorities require a provider under US jurisdiction to disclose data in its “possession, custody, or control”, wherever that data is stored source.

For Evanesca's models, the company that runs the model is European: OVHcloud, Scaleway, Mistral AI or Infomaniak, none with a US parent, so a US order cannot reach those requests directly — or AKI.IO, a German company whose shareholders are not published. OVHcloud and Scaleway each have a US entity, which is why Evanesca says “directly”; the sovereignty page explains it, and the sources page quotes the law.

Evanesca does not use OpenRouter: it serves only models run in Europe, each by one European operator, and the passport under each answer names it. See the models.

Where OpenRouter is the better choice

• Breadth. Hundreds of models, including closed ones from OpenAI, Anthropic and Google, which Evanesca does not serve at all source.
• Price on common open-weight models: the cheapest host sets the headline price, and a “Floor” routing mode goes after it source.
• Failover. When a host returns an error, OpenRouter can move the request to another host of the same model, as far as your filters allow source. Each of Evanesca's models has a single operator.
• Controls for larger accounts: data-policy filters, zero-data-retention routing on every plan, EU or US in-region routing, and SSO on Enterprise source.
• It is running today. Evanesca has not opened yet.

If you need the closed models Evanesca does not serve, or the lowest price on a widely hosted open model, OpenRouter does that and Evanesca does not.

Where Evanesca fits better

• A French company, and a European operator for every model sovereignty.
• Europe only, with no exception: no model outside Europe, not even as an option, and no fallback outside Europe sovereignty.
• One price per model, VAT included, and no subscription: credit from €2, and credit that never expires pricing.
• A passport under every answer in the chat: which model wrote it, who ran it, in which country, under which law, and what was kept passport.
• A chat with an Agent mode on the same models, for colleagues who don't write code Agent mode.
• The same API shape: OpenAI-compatible chat completions, model list and embeddings, with embedding models that also run in France docs · embeddings.

None of this makes Evanesca cheaper or broader. It makes the route of a request short, fixed and stated.

Using both

Both speak the OpenAI API, so moving a workload means changing the base URL and the key. A common split: requests that carry personal, client or regulated data go to Evanesca's models; experiments, evaluations and the closed models Evanesca does not serve go through OpenRouter.

from openai import OpenAI

# Models run in Europe: France, Mistral AI's EU region, Switzerland
europe = OpenAI(base_url="https://evanesca.app/api/v1", api_key=EVANESCA_API_KEY)

# Everything else, across many hosts
wide = OpenAI(base_url="https://openrouter.ai/api/v1", api_key=OPENROUTER_API_KEY)

What matters is that the route for sensitive data is decided in your code, not by whichever host is cheapest that day. The API documentation lists Evanesca's endpoints.

How this page was checked

OpenRouter's pages were read on 6 October 2026; prices, terms and documentation change, so that date matters. Evanesca's prices come from its catalogue, and every price verdict on this page is computed from those figures. The dollar is converted at the ECB reference rate of the same day. If a line is out of date, write to us and we will correct it.

• OpenRouter — Llama 3.3 70B: hosts and prices
• OpenRouter — pricing, plans, fees and taxes
• OpenRouter — data collection
• OpenRouter — provider logging
• OpenRouter — in-region routing
• OpenRouter — model list
• OpenRouter — models eligible for EU routing
• OpenRouter — EU endpoints for GPT-OSS 120B
• OpenRouter — terms of service, updated 31 August 2026
• AWS Customer Agreement

Frequently asked questions