Paris-based startup Weflayr has closed a 2.5 million dollar pre-seed round to build an intelligent layer for routing AI inference, the process of running models to answer real user requests. The round was led by 13books Capital, with participation from Founders Future, Kima Ventures, Plug and Play, Station F Ventures and a General Catalyst scout, among others.
The problem Weflayr targets is familiar to any team juggling today’s model market. Hundreds of models now compete on quality, speed and price, yet most companies rely on only a few, because traditional gateways route requests without knowing how each model performs on a specific workload. Weflayr instead benchmarks models against a customer’s own production data, lets teams approve trusted models and set routing rules, and matches each request to similar calls it has already measured.
That approach also ties inference choices to costs and margins across customers and features, an increasingly important link as AI bills grow. In its first deployment, the company says it cut inference costs by more than 55 percent at equivalent quality, and delivered 15 percentage points higher relative quality at the same cost compared with a frontier model.
Weflayr was founded by Romain Mazeries, former chief executive of Mangopay, and Noé Achache, former head of AI engineering at Theodo Data & AI. The funding will expand its engineering and AI research teams and support growth in Europe and the United States. For teams comparing models, hubs such as Hugging Face and assistants such as ChatGPT and Claude show just how crowded, and how expensive, the model landscape has become.
Source: The AI Insider
