Weflayr raises $2.5m pre-seed to route AI calls to the best model

Weflayr raises $2.5m pre-seed to route AI calls to the best model

22 October 2026: Weflayr, a Paris start-up building a routing layer for AI inference, has raised $2.5 million in pre-seed funding led by 13books Capital, with Founders Future, Kima Ventures, Plug and Play, Station F Ventures, Roxanne Varza as a General Catalyst scout, and angel investors. The money pays for engineering and AI research hires, product development and expansion across Europe and the United States.

Not yet independently verified. The round rests on Weflayr's own announcement, its cost and quality results come from a single deployment it describes, and it names Station F rather than a registered address as its base. We will update this when it can be confirmed, and remove this note.

Routing on your own benchmarks

AI gateways already sit between an application and several model providers and switch requests between them. Weflayr's argument is that most do so blind, on price or on generic leaderboards, so companies stick to a handful of large models because quality cannot be risked. Its product benchmarks candidate models against a customer's real production calls, scored against the customer's own quality and cost goals. The customer approves which models may be used, and Weflayr then matches each incoming request to similar calls it has already benchmarked and sends it to the best-performing approved model within those limits. It also links each inference decision to its business context, so costs and margins show up per customer, feature and workflow.

In its first deployment, Weflayr says, inference costs fell by more than 55% at equivalent quality, and relative quality rose 15 percentage points against a frontier model at the same cost. Those are its own figures from one customer.

From payments routing to model routing

The company was founded by Romain Mazeries, formerly chief executive of the payments company Mangopay, and Noé Achache, formerly head of AI engineering at Theodo Data & AI, and works from Station F in Paris. The founders draw the parallel themselves: payments learnt to route each transaction to the best processor, and they expect inference to follow at greater speed. "There is a massive opportunity for an independent intelligence layer," said Mike McFadgen, partner at 13books Capital.

Weflayr is onboarding its first customers and plans wider availability in the coming weeks; for now there is a waitlist.

A crowded layer

Buyers weighing a gateway should ask how a router's benchmarks are kept current as providers update models under the same name, since a routing decision is only as good as its last evaluation. Our AI gateways category lists two products and our API management category 33, several of which now route model traffic too.

Company profile on TrustList: Weflayr

Sources

• Weflayr: We just raised $2.5M to reinvent AI inference routing, 22 October 2026