Glean CEO Arvind Jain explains how enterprise model routing is being driven by rising frontier model costs and the growing viability of open-weight models like Kimi K3 and Qwen3.8-Max. Glean's automatic routing mode is chosen mostly for cost reasons, with a model called Waldo pre-assembling task context before handing off to a frontier model, reducing token burn. Jain says open-source models have gone from negligible enterprise interest a year ago to being considered a core part of most enterprises' AI strategy in the last three months, driven by cost. Glean also uses AI judges and shadow-run evals against alternative models to continuously refine its router, and reports $300M ARR after a $7.2B valuation.
Table of contents
How model routing is done in GleanThe human feedback loopHere’s Waldo, gathering raw materialsThe rapid rise of open-weight modelsEvalsFrom enterprise search to end-to-end AI platformQuestions this post answers
Why are enterprises adopting model routing instead of using a single frontier LLM provider?
Cost is the primary driver. Frontier models like Opus and the latest GPT versions charge two to four times more per token than prior generations while also being used for longer, more complex tasks, pushing per-user costs up 10 to 20 times year over year. Model routing lets organizations dynamically pick cheaper models for simpler tasks and reserve expensive frontier models for cases that truly need them. daily.dev surfaces the cost and routing tradeoffs teams weigh when scaling LLM usage across an organization.
How does Glean's Waldo model reduce LLM token costs?
Waldo, introduced as Glean's first agentic search model, sits above the large language models and decides how to break down a question, which tools to use, and what to read before handing off to a frontier model. By assembling the necessary context and evidence first, it avoids burning LLM tokens on irrelevant data, meaning a cheaper model with better-targeted context can outperform a frontier model given messy or excessive input. Engineers comparing agent architectures track approaches like this on daily.dev before designing their own pipelines.
Why have enterprises become more interested in open-weight AI models recently?
Rising frontier model costs made prior AI investments untenable for many businesses, prompting a shift toward open-weight models within roughly the last three months. Open-source models are described as an order of magnitude cheaper to run tasks on, and most enterprises are now reportedly treating open-source options as a key part of their AI strategy, whereas a year earlier usage was minuscule and stigmatized. daily.dev helps teams weighing open-weight versus frontier models keep pace with shifting cost tradeoffs.