Glean CEO on Latent Space: Routing Puts Tasks at $0.45 Versus $1.84, and Enterprises Went From Near-Zero Open-Weights Usage to Treating It as Strategy in Months
Richard MacManus interviewed Glean CEO Arvind Jain about model routing as an enterprise cost-control layer, with three modes (employee choice, admin control, automatic dynamic routing) where customers pick automatic almost entirely for economics. Co-founder Tony Gentilcore claims the platform is 4x more cost-effective than Claude Code at $0.45 per task versus $1.84, and Glean's Waldo agentic search model cuts latency 50% and tokens 25% by filtering queries before they hit a frontier model. Jain's sharper claim is about the open-weights shift: 'open source is an order of magnitude cheaper to do tasks,' and enterprise customers who used almost none of it months ago now call it a key part of their AI strategy, with 'nobody willing anymore to rely on only one model provider.'
Source
↳ Follow the thread