You.com has launched Inference Router, an OpenAI-compatible API that routes production requests across 12 open-weight models with a 99.9% minimum uptime SLA, zero data retention, and 100% US hosting — all with no platform fee.
You.com Inference Router (Credits: You.com)
The router sits at api.you.com and offers both OpenAI-compatible and Anthropic-compatible routes, so existing SDK code migrates by changing only the base URL and API key. Requests hit You.com’s own Inference Direct capacity first, then fail over to vetted US-hosted partners — Parasail, Wafer, Alibaba Cloud, Meta and Inception, with Prime Intellect coming soon.
You.com Inference Router: 12 Open-Weight Models With 99.9% Uptime SLA
The lineup runs from GPT-OSS 120B (125K context) to Kimi K3 (a 2.8T-parameter model) and Qwen3.8 Max (2.4T MoE, 1M context). Prices are published per million tokens — GPT-OSS 120B at $0.097/$0.73 — and each model carries one published price regardless of which partner serves the request.
You.com is pitching the product at a familiar pain point: dev teams test open-weight models on aggregators, then stall in security review. The router answers reviewers with zero data retention (prompts and outputs processed in memory, never stored), SOC 2 compliant infrastructure, DPAs on request, per-key spend caps, and daily usage views broken down by model, cost and provider.
One headline feature is server-side web search: developers declare a web_search or web_extract tool and the router runs searches for the model on the same server — up to 20 results per search and a default of 30 search rounds, returning citations. You.com says internal tests show up to 15% lower latency from co-locating search with inference.
Open-Weight Models Match Frontier Accuracy at Lower Cost, You.com Claims
In a self-reported LiveNewsBench run of 1,329 questions, You.com says GLM-5.3 paired with its search scored 86.53% at $0.0485 per question — matching GPT-6.1 Sol with built-in search on accuracy while costing roughly half of Claude Sonnet 5.5’s $0.0920. DeepSeek V4.1 Flash with You.com search trailed by 2.26 points but cost about 77% less per question.
As always, treat vendor benchmarks as marketing until independent labs confirm them — but the message is clear: open-weight models plus grounded search are closing the gap with frontier systems on real-world knowledge tasks.
Also Read: IBM Releases Granite Speech 5.0-2B: Open Multilingual Transcription and Translation Model
Source: MarkTechPost — read the full story there.
