Comparison
Frontière AI vs OrcaRouter: a zero-markup router, or models hosted in Europe
OrcaRouter and Frontière AI both put many models behind one OpenAI-compatible API, and on price OrcaRouter wins. It bills tokens at the upstream provider's list price with no markup — on DeepSeek V4 Flash 0731, €0.19 ($0.22) per million input tokens before tax, against our €0.56 before tax (€0.672 VAT included) — and charges for team features instead. The company behind it, Continuum AI Pte. Ltd., is incorporated in Singapore; its privacy policy lists servers in Singapore, the United States and Japan, and its region setting tags compliance reports without moving traffic. Frontière AI is a French company: its 17 sovereign models run at EU providers with no non-EU capital control, the others are labeled fast access (US) before you call them, and it stores no prompts at all. If the lowest token price and the widest catalogue, closed models included, are what you need, OrcaRouter is the stronger offer. If your data has to be processed in Europe by a named European provider, that is what Frontière AI is built for.
Side by side
Every row below comes from the other side's own published pages, checked on 2026-10-13 and linked at the end of this section — not from a third-party comparison.
| Criterion | Frontière AI | OrcaRouter |
|---|---|---|
| Operating company | FLEECE AI SASU | Continuum AI Pte. Ltd. |
| Headquarters | France | Singapore (registered office in its terms; launch press releases datelined San Francisco) |
| Governing law | French / EU law | Laws of Singapore; Singapore courts and SIAC arbitration (its terms) |
| US CLOUD Act exposure | Inference on the 17 sovereign models runs at EU companies; requests transit the gateway on Vercel (US company, Paris region, no prompt or response stored) | Not a US company, but its privacy policy lists infrastructure in Singapore, the United States and Japan at a cloud given as “e.g., AWS, GCP”, with Cloudflare in front of the gateway |
| How the catalog works | Curated catalog, 24 models live, each with a named primary provider | Router across eleven upstream providers (OpenAI, Anthropic, Google, DeepSeek, Qwen, Kimi, Z.ai and others) at their list price, with an MIT-licensed self-hosted edition |
| DeepSeek V4 Flash 0731, per 1M tokens (in / out) | €0.672 / €1.344 incl. VAT (€0.56 / €1.12 excl. VAT) | €0.19 ($0.22) / €0.571 ($0.66) excl. tax (our reading) |
| Prompt storage | Never stored by Frontière AI — only token counts and the amount charged. Sovereign providers: OVHcloud zero retention; Scaleway none except error or abuse requests, up to two weeks; fast-access models: the serving provider's own policy | Not stored unless you turn request logging on, then kept 30 days by default and up to 180 (its docs); request metadata, source IP included, kept 13 months |
| Jurisdiction visible per model | Yes — sovereign / fast access, on the card and in the API response | The workspace region tags compliance reports and “doesn't yet pin where your data is stored or where inference runs”; in-region processing is a separate option set per model through its sales team |
OrcaRouter publishes in US dollars, shown converted at 1 USD = 0.8655 EUR (rate of 2026-08-05) next to the original figure. OrcaRouter's terms say taxes, where applicable, are the customer's responsibility unless it is legally required to collect them, and its pricing page gives provider list rates in USD without mentioning VAT: we read them as before tax (checked on 2026-10-13). Frontière AI publishes in euros, margin and VAT included — the amount actually debited — so the price verdicts on this page compare both sides excluding VAT (our price ÷ 1.2, French VAT).
Why the price gap? OrcaRouter passes the upstream list price through and earns on its paid plans; it lists this model under DeepSeek, and its docs do not say which host serves its DeepSeek models. Frontière AI serves this model through Scaleway, an EU sovereign provider, at one price with a fixed margin. Part of the gap is that margin, part is the route: a European provider named in advance, rather than whichever upstream the router reaches. Excluding VAT on both sides (Frontière AI: €0.56 in / €1.12 out per million), on this model OrcaRouter is cheaper on both input and output.
Sources, checked on 2026-10-13: api.orcarouter.ai — models & prices (DeepSeek V4 Flash 0731) · orcarouter.ai — pricing · orcarouter.ai — terms · orcarouter.ai — privacy policy · docs.orcarouter.ai — data handling · docs.orcarouter.ai — data residency (report tag) · docs.orcarouter.ai — in-region processing · github.com — OrcaRouter Lite (MIT)
Where OrcaRouter is the better choice
- Zero markup on tokens: you pay the upstream list price, and its plans charge for team features rather than tokens (Hacker free, Team $499 a month, Enterprise on quote).
- A far wider catalogue — about 200 models, closed ones from OpenAI, Anthropic and Google included — with an automatic router (orcarouter/auto) and failover between upstreams.
- A self-hosted edition, OrcaRouter Lite, published under the MIT licence, if you would rather run the router yourself with your own provider keys.
- Google GenAI SDK compatibility on top of the OpenAI and Anthropic formats. (Frontière AI takes the OpenAI and Anthropic formats.)
Where Frontière AI is the better choice
- The operating company is French. There is no US parent, licensee or affiliate through which a US order could reach your data on the sovereign models.
- Europe is a default, not a negotiation: each sovereign model is served by a named European provider for every customer, from the first call. On OrcaRouter, the self-serve region setting does not move traffic, and in-region processing goes through its sales team.
- Every model carries its jurisdiction before you call it, in the catalog and in the API's own response — you never have to reconstruct who served a given request.
- API prompts and completions are never written to our database — only the token counts billing requires. The web console follows the same rule: conversations stay in your browser.
How to choose, concretely
Both speak the OpenAI format, so trying one or the other is a base_url and a key:
- Lowest price per token, closed frontier models, quick prototyping: OrcaRouter.
- Personal or client data that must be processed in Europe by a company you can name in your records: Frontière AI's sovereign models.
- You want to run the router yourself: OrcaRouter Lite, under the MIT licence, with your own provider keys.
FAQ
Is OrcaRouter a European company?
No. Its terms name Continuum AI Pte. Ltd., a Singapore private limited company, under Singapore law with arbitration in Singapore, and its privacy policy says the service is operated from Singapore, on infrastructure in Singapore, the United States and Japan. Its launch press releases are datelined San Francisco.
Can OrcaRouter keep my data in the EU?
Not by default. Its docs say the workspace region tags your compliance reports and does not yet pin where data is stored or where inference runs. Processing restricted to one region exists as a separate option, set per model and arranged through its sales team. On Frontière AI, each sovereign model is served by a named European provider for every customer, and its label is shown before you call it.
Is OrcaRouter cheaper than Frontière AI?
Yes, on the model we compared. On DeepSeek V4 Flash 0731, OrcaRouter lists €0.19 ($0.22) in / €0.571 ($0.66) out per million tokens, read as before tax, against €0.672 / €1.344 VAT included on Frontière AI. Excluding VAT on both sides (Frontière AI: €0.56 in / €1.12 out per million), on this model OrcaRouter is cheaper on both input and output. It passes the upstream price through; we add a fixed margin and serve this model from a named European provider. The argument for Frontière AI is where the data goes, not the price.
Does OrcaRouter store prompts?
Its privacy policy says it does not log, store or retain prompts or outputs; its newer docs say content is not stored unless you turn request logging on, and then kept 30 days by default, 180 at most. Request metadata, source IP included, is kept 13 months, and zero data retention requires a signed agreement. Frontière AI never writes API prompts or completions to its database — only the token counts billing needs.
See what we actually serve
24 models live, each with its publisher, licence, hosting jurisdiction and real price.