You.com Launches Inference Router for Open-Weight AI Models

The Inference Router provides access to 12 live open-weight models, ranging from GPT-OSS 120B with a 125K context window to Kimi K3 and Qwen3.8 Max, which support up to a 1 million context window. Kimi K3 is a 2.8 trillion parameter model, and Qwen3.8 Max is a 2.4 trillion parameter Mixture-of-Experts (MoE) model.

The service includes a 99.9% minimum uptime Service Level Agreement (SLA) and implements zero data retention (ZDR). It also features 100% US hosting and server-side web search capabilities, operating on a pay-as-you-go model without any platform fees.

Requests are initially routed to You.com's own inference capacity, Inference Direct. If this capacity is unavailable or full, requests automatically fail over to vetted, 100% US-hosted partners including Parasail, Wafer, Alibaba Cloud, Meta, and Inception, with Prime Intellect expected soon. This failover process does not affect billing, as each model has a single published per-token price.

You.com frames the Inference Router as a solution to stalled migrations of open-weight models into production. The service addresses security review concerns regarding zero data retention, US hosting, access controls, and data processing agreements (DPAs), which dedicated providers typically require large commitments to meet.

Performance benchmarks on You.com’s LiveNewsBench indicate that open-weight models combined with You.com search can match frontier models at a lower cost. For example, GLM-5.3 with You.com search achieved 86.53% at $0.0485 per question, compared to Claude Sonnet 5.5 at $0.0920 per question, according to self-reported data.