Weflayr Raises $2.5M To Cut AI Inference Costs 55%

Paris-based Weflayr raised a $2.5 million pre-seed round led by 13books Capital to build an AI inference-routing layer it says cut model costs more than 55% at equivalent quality in its first deployment.

Trace Cohen
By Trace Cohen · Value Add Pulse AI Desk
Founder & Editor-in-chief, Value Add VC
2 min read
Weflayr Raises $2.5M To Cut AI Inference Costs 55%: key figures
ShareXLinkedInEmail

THE RUNDOWN

1

AI inference routing is becoming its own infrastructure category as companies juggle dozens of models and need automated, cost-aware switching.

2

Weflayr's early numbers -- a 55%+ cost cut and a 15-point quality gain versus a single frontier model -- are self-reported from one deployment, not independently audited.

3

The round is tiny and the company is pre-scale; the real test is whether 13books Capital's backing attracts US customers, not just European ones.

4

Founders Romain Mazeries (ex-Mangopay CEO) and Noe Achache (ex-Theodo Data & AI) bring payments and applied-AI engineering backgrounds, an unusual combination for an infra startup.

The VC Read

Value Add VC analysis

Watch the retention number, not the cost-cut number: a 55% inference-cost reduction is easy to show in a pilot and hard to sustain once a customer's traffic mix changes. The diligence item is whether Weflayr's routing holds its savings after three months of real production traffic, not the demo deployment 13books Capital just funded.

Analysis

Weflayr, a Paris-based startup, closed a $2.5 million pre-seed round led by 13books Capital, with Founders Future, Kima Ventures, Plug and Play, Station F Ventures, General Catalyst scout Roxanne Varza and a group of angel investors also participating.

The product lets AI companies benchmark models against their own production data, approve trusted models, set routing rules, and match each new request to similar previously benchmarked calls -- tying inference decisions directly to cost and margin across customers and features rather than leaving model choice to a single default.

Why inference routing is becoming its own category

As companies run workloads across OpenAI, Anthropic, open-weight models and more, picking the right model per request rather than per application has become a cost and quality problem of its own. Weflayr says its first deployment cut inference costs more than 55% at equivalent quality and delivered roughly 15 percentage points higher relative quality at the same cost compared with routing everything to a single frontier model. 13books Capital partner Mike McFadgen framed the opportunity as building an independent intelligence layer at a moment when AI inference is becoming critical infrastructure in its own right.

Weflayr is entering a field that already includes OpenRouter, which routes requests across dozens of hosted models, and smaller specialists like Martian and Not Diamond that also sell model-selection logic as an API. The risk for a $2.5 million pre-seed company is that this category attracts well-funded entrants quickly once a buyer pattern is proven, and Weflayr's single-deployment numbers have not yet been tested at the multi-customer scale its larger competitors already operate at. However, the founding team's background is a point in its favor: Romain Mazeries, Mangopay's former CEO, ran payments infrastructure at scale, and Noe Achache, previously head of AI engineering at Theodo Data & AI, has applied-AI delivery experience most infra-only founders lack.

The pair are using the round to expand engineering and research and push into the US market, where most of the routing competition already sells -- a harder market to break into than the European seed-stage circles that just backed them.

ShareXLinkedInEmail

Key Sources

1 source

THE WIRE in your inbox— Tech, startup & VC news with The VC Read, a few times a week. Free to subscribe, no spam.

No spam, unsubscribe anytime. Sent via Substack · Privacy · or subscribe on Substack ↗