You.com

Kimi K3.1 Pro is live on Inference Router from day 0. K3.1 Pro is Kimi (Moonshot AI)'s most capable open-weight model yet, with 1M context window, and a reported 65.7% on Terminal-Bench 4.0, within a point of the strongest closed models. It’s built for long-horizon coding and agentic work. On Inference Router, K3.1 Pro comes with: • At least 99.9% uptime SLA, with automatic failover to vetted partners • US hosting, with ZDR available on every call • Pay-as-you-go, 0% platform fee Try it in the playground or talk to an engineer. [link]

Kimi K3.1 Pro is live on Inference Router from day 0. K3.1 Pro is Kimi (Moonshot AI)'s most capable open-weight model yet, with 1M context window, and a reported 65.7% on Terminal-Bench 4.0, within a point of the strongest closed models. It’s built for long-horizon coding and agentic work. On Inference Router, K3.1 Pro comes with: • At least 99.9% uptime SLA, with automatic failover to vetted partners • US hosting, with ZDR available on every call • Pay-as-you-go, 0% platform fee Try it in the playground or talk to an engineer. https://lnkd.in/g9KGzmFY

Excited to partner with Fireworks AI to deliver real-time web search for open models across its Frontier Ecosystem. Point any model at the You.com MCP server, and it searches the web whenever needed. fireworks.ai/partners

Beam from Reflection is now available on Inference Router. Beam is a US-built open-weight model for coding and agentic tasks. Now you can run it in production without a big upfront commitment. On Inference Router, Beam comes with: • US hosting, with ZDR available on every web search call. • Pay-as-you-go, 0% platform fee, volume discounts as you scale. Try Beam in the playground or talk to an engineer. https://lnkd.in/g6JEuv7D

Swapping models won't fix your accuracy if search is the bottleneck. Your evals are measuring the retrieval layer as much as the model. Zairah Mustahsan on running evals you can trust, from The AI Conference. https://lnkd.in/gj9VWc_n

Saahil J.: "One thing for sure is that agents will be performing more searches than humans for the foreseeable future." Saahil joined You.com as a founding engineer in 2020, when it was a consumer search engine - now he's CTO, building search for AI agents. BrightEdge's April data already put AI agent requests at 88% of human organic search activity - and it's vastly increased since then. All of our infrastructure needs to adapt to the agentic era, including search. • YouTube (full video): https://lnkd.in/ggUPg5PT • Episode page (chapters + transcript): https://lnkd.in/gJUbjNBT • Apple: https://lnkd.in/gC-wDhKF • Spotify: https://lnkd.in/gW9mHBKD Brought to you by: • Walrus Foundation, presenting sponsor of Season 4. Walrus Memory gives AI agents portable, verifiable memory that carries context across apps, sessions, and other agents: https://walrus.xyz/cot • Coral Bricks AI sponsored this episode. CoralBricks serves open models through one API, fast and at low cost. Get $5 in credits on signup, plus another $5 on your first purchase: https://lnkd.in/gp8UZbTE • Inngest, presenting sponsor of Season 4. Durable execution for agents in production: https://lnkd.in/gZgSJf3v • WisdomAI sponsored this episode. In WisdomAI's latest benchmark, Claude Code alone topped out at 73% accuracy, and WisdomAI closed the gap: https://lnkd.in/gxkcpCEk 4 takeaways: 1. Routing between models is increasingly important: At DevDay, OpenAI and Baseten announced that enterprises can spend OpenAI commitments on open models in Codex. My read: OpenAI wants the whole enterprise model budget. And as Saahil notes, cheap open models are expanding the opportunity for harnesses, because routing saves money at the same accuracy. 2. Agents search differently than people: They can run a thousand searches at once, and with 10 results in hand the order matters much less to them than to us. Saahil: "Recall generally seems to matter more than precision when it comes to agent results." 3. We built the self-driving cars before the roads(recurring theme for our episodes you might notice!): Models are often capable enough; the tools and infrastructure around them lag, and search is one of those roads. If you remember our conversation with Yutori about browser use and their navigator models, APIs and other infra are in similar boats. 4. Judge search by whether the agent answered: Saahil's test: "Was it able to answer the question? Did it do something useful?" Static benchmarks can be memorized, so build ones that change over time, and increasingly, develop custom evals for/with partners. Thank you for joining me on the Chain of Thought Podcast, Saahil! #AIpodcast #chainofthoughtpodcast #ai #aisearch

Midterms are tomorrow. Your model's training data doesn't know that. So we asked our Web Search API with Knowledge. See the latest odds below, or run the query yourself👇 https://lnkd.in/g9mNgtiN

We're partnering with Fireworks AI to give open models the live web. Kimi, DeepSeek, GLM, Qwen, MiniMax, gpt-oss: point any of them at the You.com MCP server and they search when needed. No key to start. 100 free queries a day. More: https://lnkd.in/gHyJBQfw

Adding web analytics data to the You.com Web Search API, via Knowledge. Your agent gets market share across browsers, operating systems, search engines, devices and more from StatCounter, returned alongside web results in a single call. Web analytics joins stocks, crypto, Treasury, rent, rates, prediction markets, sports, and weather in Knowledge. Same $5 CPM, no extra setup. Try it yourself: https://lnkd.in/gNq_ebuj

SCIRA AI now runs on the You.com Web Search API by default. Every web-mode answer on Scira, across 5M+ searches, now cites from several extracted passages per page. Fresher news, date-bounded Lookouts, and images in the same response. More: https://lnkd.in/g-xjMdZJ