Why AI Agents Need More Than One Model

Intelligence isn't one-size-fits-all. AI agents are built with many models, each bringing different strengths to the work. Some tasks call for deep reasoning. Others demand speed, lower costs, domain expertise, or stronger data control. Frontier models provide state-of-the-art reasoning for the hardest tasks. Open models can be fine-tuned on company data to excel at specialized work. They can also run on-premises, helping keep sensitive data secure while reducing latency for high-volume tasks. Glean uses this system-of-models approach to help enterprises search and act on their data. Waldo, a specialized model post-trained on NVIDIA NEMOTRON® 3.0, gathers context across sources like support tickets, slack, and survey data. With that context, it decides whether the agent should respond directly or hand off to a frontier reasoning model for deeper analysis. For simple questions, Waldo delivers all the context to a model that returns the answer in one shot. The response is fast. If the question is more complex, Waldo sends the task to a frontier model that reasons deeper. Because the frontier model receives the right context up front, it can produce structured analysis with themes, evidence, and segment breakdowns more efficiently. Routing lets Glean search enterprise context 10 times faster. This translates to 50% lower latency and 25% fewer tokens, with no reduction in answer quality. In a system-of-models, frontier and open models work together to make your agent smarter, faster, and more efficient.