You.com Inference Router gives developers 12 AI models, most of them open-weight, through one OpenAI-compatible API, and You.com puts an uptime guarantee of at least 99.9 percent behind it, with service credits as the remedy. The lineup includes Moonshot AI's Kimi K3, Z.ai's GLM 5.3, DeepSeek V4 Pro, Alibaba's Qwen3.8 Max, MiniMax M3, Inception's Mercury 3, OpenAI's GPT-OSS-120B and Meta's Muse Spark 1.3 and Muse Wonder, plus Flash variants from Z.ai, DeepSeek and Alibaba. A thirteenth model, Beam from Reflection, is marked as coming soon.
The router sends traffic first to You.com Inference Direct, the company's own capacity. If that is full or down, requests fail over to partner providers: Parasail, Wafer, Alibaba Cloud, Meta and Inception, with Prime Intellect listed as coming. You.com says every provider in that chain is hosted in the US and meets its compliance requirements, including zero data retention, and that it does not store prompts or outputs on any model call or web search request across its managed stack. Accounts run on SOC 2 compliant infrastructure, and enterprise customers can add a data processing agreement. Failover, the company adds, never changes the bill.
The quickstart shows more of how it works. Each model reports its token limits where a limit holds for every request. Access is set per account, so two keys can see different model lists. You.com Inference Router serves an Anthropic-compatible /v1/messages endpoint next to the OpenAI-style /v1/chat/completions one, so either SDK family works after a swap of the base URL and key. A request that declares a web_search or web_extract tool has the router run the searches and page fetches and return the answer with its sources.
Each model has one per-token price with no platform fee on top, and volume discounts and higher throughput come with larger deployments. On the models page, Kimi K3 has a 1M-token context and costs $2.85 per million input tokens and $14.25 per million output tokens. Input prices there run from $0.097 per million tokens for GPT-OSS 120B to $4.00 for Meta's Muse Wonder. In its comparison table, You.com lists OpenRouter with no published uptime SLA and a 5.5 percent platform fee.
The performance figures come from You.com. Running web search on the same server as the model cuts latency by up to 15 percent in its internal testing. A Braintrust evaluation the company cites found an open-weight model paired with You.com search matched or beat a leading closed model using its own built-in search on questions that need current information, at up to 75 percent lower cost per question. You.com notes that results vary by model, task and evaluation setup.
An existing OpenAI client moves over by pointing its base URL at https://api.you.com/v1 with a You.com API key, then calling a model ID such as kimi-k3. Beam and Prime Intellect are the two additions You.com lists as still to come.
