The frontier is leaving the chat — Aug 26

Four Aug. 25 signals show frontier AI leaving the chat window: measured inference silicon, physics-first models, agents inside law-firm systems, and capital for keeping acting systems inside policy.

OpenAI put numbers on Jalapeño. At Hot Chips, OpenAI shared the first InferenceX results for its custom inference chip. Across GPT-OSS 120B, DeepSeek R1, and Kimi K2.5 1T, OpenAI reports 1.5–1.9× more AI work per watt and 1.7–3.6× lower end-to-end latency than comparison systems. First deployments are planned for late 2026 in small volumes.12

Accelerated Understanding is betting on physics, not text. Anima Anandkumar and Benedikt Jenik launched a company around neural operators after declining a Bezos-backed Project Prometheus offer. The firm says its model handled 5 trillion pieces of data in a single prompt in tests, and it is aimed at chip design, robotics, weather, and geological work rather than chat.3

Google is putting Gemini agents into legal stacks. Gemini Enterprise for Legal launched in preview with domain skills, MCP connectors into systems such as iManage, NetDocuments, Everlaw, Harvey, and Legora, and collaborating firms including Cleary Gottlieb, Freshfields, Weil, and Williams & Connolly.45

Alice raised $140M to test and contain acting models. The former ActiveFence company says the round, led by Apax Digital, brings total funding to $280M. Alice works with Anthropic, Google, and Cohere, claims work with 8 of the 10 leading model labs, and says it is approaching $100M ARR.6