Glean Waldo: An agentic search model that pairs with LLMs to deliver frontier intelligence with ~50% lower latency and ~25% fewer tokens

What this post added

Introduced Waldo, an agentic search model that acts as a specialized front-end to frontier LLMs. Waldo is trained to break down queries, select tools, and gather evidence before handing off to a frontier model, significantly reducing latency (~50%) and token usage (~25%). The training process involves Direct Preference Optimization (DPO) on tool-use patterns and reinforcement learning with outcome-based rewards. Waldo's execution also informs adaptive reasoning, dynamically determining the required reasoning level for a task.