Woofun AI notes that enterprise spending on open-source large models currently represents only 11% of total expenditure, as most organizations remain in early exploratory phases relying on closed-source solutions. Jesse Zhang, CEO of Decagon, indicates that this dynamic will shift as application scenarios mature, with open-source models expected to dominate production environments owing to superior latency and fine-tuning capabilities.

Decagon has already migrated 90% of its call volume to open-source weight models, driven by the need for interaction speed and customization rather than cost reduction. Zhang predicts a future division of labor where top closed-source labs lead new field exploration while open-source models handle mature business production, a transition expected to unfold over several years.