Evanesca vs OrcaRouter: a zero-markup router from Singapore, or models run in Europe

At a glance

Each OrcaRouter cell links to the OrcaRouter page it comes from; every one was read on 13 October 2026. Each Evanesca cell links to the Evanesca page that describes it.

Price: OrcaRouter costs less

OrcaRouter lists several of the open-weight models that Evanesca serves. The table compares them at the prices of its public model list on 13 October 2026; OrcaRouter's own id for each is given next to its price. On Evanesca, DeepSeek V4 Flash is the 0731 build that Scaleway runs in Paris.

OrcaRouter 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.

OrcaRouter is cheaper on input and on output on every row of the table. On DeepSeek V4 Flash, OrcaRouter costs 65% less on input and 48% less on output. The gap is not a discount: OrcaRouter says it bills “at provider price” and adds nothing per token source, and its terms charge “the published cost price of the upstream provider” source. Its paid plans sell team features, not cheaper tokens.

Two details sit next to that price. Bringing your own provider key is described as free “during the introductory period” on one page source, while the billing page speaks of “a small platform fee” source. And OrcaRouter lists its DeepSeek models as owned by DeepSeek, without saying, in the pages we read, which company hosts them or where.

Evanesca's price is fixed model by model: what the European operator charges, a margin and VAT. It is higher, and it pays for a route that does not change from one request to the next.

What happens to your prompts

OrcaRouter's privacy policy makes a strong promise: “we do not log, store, or retain the content of your prompts or model outputs” source. Its documentation, more recent in what it describes, adds the exception: bodies are not stored “unless you turn request logging on for your workspace”, and then they are kept 30 days by default, up to 180 source. Whoever administers the workspace decides.

« Prompts aren’t logged unless you turn logging on. Sign a zero-retention agreement to lock it off. »

Some data is kept in every case: one metadata row per request — time, model, token counts, latency, status and the source IP address — for 13 months source. A zero-data-retention agreement locks logging off and routes only to zero-retention upstreams; it is signed per workspace, through sales, and does not remove the metadata source. Then the request leaves for the upstream provider, “under their own terms and privacy policy” source. OrcaRouter's terms say it does not train any model on your content source.

On Evanesca, nothing about prompts is a setting. Evanesca stores no prompt and no answer, and each model's operators are fixed in advance — the first choice and, for some models, a European fallback — with the rule of the one that answered printed on the passport under each answer: Scaleway may keep and read requests that cause errors or look abusive, for up to 14 days (Scaleway's data privacy page). Mistral AI keeps API inputs and outputs “for thirty (30) rolling days to monitor abuse” (Mistral's privacy policy). The passport page gives every operator's rule.

A region tag is not a region

OrcaRouter lets a workspace declare a region — US, EU, UK, Asia-Pacific, China or global. Its documentation is candid about what that does: it stamps your signed compliance reports with the region and stores them there, and “setting region: eu does not route your model traffic through EU infrastructure” source. Upstream providers, the same page says, process requests in their own regions.

Keeping processing inside a region is a separate feature. It is set per model, “because not every model is offered by its provider in every region”, it covers processing but not the metadata, and it is obtained through sales source. Until then, a request is served “wherever the model you asked for is available”.

The company and its own servers sit outside the EU. The contracting party is CONTINUUM AI PTE. LTD., under Singapore law source; the privacy policy says the service is “operated from Singapore”, on a cloud named only by example (“e.g., AWS, GCP”) in Singapore, the United States and Japan, and relies on standard contractual clauses for data leaving the EEA source. Cloudflare sits in front of the gateway source. Its launch release was datelined San Francisco source.

For Evanesca's models, the company that runs the model is European and named in the catalogue: OVHcloud, Scaleway, Mistral AI and Infomaniak have no US parent, so a US order cannot reach those requests directly; AKI.IO is a German company whose shareholders are not published. OVHcloud and Scaleway each have a US entity, which is why “directly” matters; the sovereignty page explains it. Evanesca does not use OrcaRouter, or any router, to reach them.

Where OrcaRouter is the better choice

• Price. OrcaRouter costs less on input and output on every row of the table, and it adds no markup to the upstream list price source.
• Breadth. Closed models from OpenAI, Anthropic and Google, video models, and many open-weight ones, behind one key source. Evanesca serves no closed model from those companies.
• An auto-router. Send the model id orcarouter/auto and it picks a model per request, with cheapest, quality, balanced and adaptive strategies source.
• Two API formats. It answers OpenAI-style requests and, at the same host, Anthropic's Messages format source.
• A self-hosted edition. OrcaRouter Lite is published on GitHub under the MIT licence, with your own provider keys source.
• It is running today, with a free plan. Evanesca has not opened yet.

If you need closed frontier models, the lowest price on a widely hosted open model, or a router you can run yourself, OrcaRouter does that and Evanesca does not.

Where Evanesca fits better

• The place is fixed. Every model runs in Europe, at a named European operator, with no option to leave and no fallback outside Europe sovereignty.
• A French publisher under French law, and operators you can look up before your first request models.
• Prices in euros, VAT included, with no subscription: credit from €2 that never expires pricing.
• A passport under every answer: the model, its operator, the country, the law, and what the operator keeps passport.
• A chat with an Agent mode on the same models, for people who don't write code Agent mode.
• An OpenAI-compatible API on the same catalogue docs.

None of this makes Evanesca cheaper or broader. It makes the answer to “who processed this, and where?” the same for every request.

Using both

Both speak the OpenAI API, so the split happens in your code: personal, client or regulated data goes to Evanesca's models; experiments, evaluations and the closed models Evanesca does not serve go through OrcaRouter.

from openai import OpenAI

# Models run in Europe, each by a named operator
europe = OpenAI(base_url="https://evanesca.app/api/v1", api_key=EVANESCA_API_KEY)

# Everything else, at upstream list prices
router = OpenAI(base_url="https://api.orcarouter.ai/v1", api_key=ORCAROUTER_API_KEY)

Other routers are built in Europe and route only to European providers; our list of European inference providers compares them with the companies that run the models themselves.

How this page was checked

OrcaRouter's pages and its model list were read on 13 October 2026; prices and documentation change, so that date matters. Evanesca's prices come from its catalogue, and every verdict on this page is computed from those figures, with the dollar converted at the ECB reference rate. If a line is out of date, write to us and we will correct it.

• OrcaRouter — public model list and prices
• OrcaRouter — pricing and plans
• OrcaRouter — terms, effective 18 April 2026
• OrcaRouter — privacy policy, updated 18 April 2026
• OrcaRouter — data handling
• OrcaRouter — zero data retention
• OrcaRouter — compliance report region
• OrcaRouter — in-region inference
• OrcaRouter — data flow
• OrcaRouter — upstream providers
• OrcaRouter — auto-router
• OrcaRouter — bring your own key
• OrcaRouter Lite on GitHub
• Launch release, PR Newswire, 8 May 2026

Frequently asked questions