Opinion: Enterprise Big Model Spending on Open Source Accounts for Only 10%, But Mature Production Environment Will Be Dominated by Open Source Models

According to DolphinBeat monitoring, public opinion often touts that open-source large models are devouring everything, but enterprise spending data presents a different picture. Jesse Zhang, Co-Founder and CEO of Decagon, an enterprise AI customer service intelligent agent platform, pointed out that the open-source share in the total expenditure of enterprise large models has now dropped to 11%. This decline is due to the fact that the AI usage of most enterprises is still in the early exploratory stage and therefore defaults to relying on closed-source models. However, he emphasized that once the application scenarios mature, open-source models will take over the production environment with extremely low latency and fine-tuning advantages.

In Decagon's own production environment, 90% of the call volume has already switched to open-source weight models. The driving force behind this transformation is interaction speed and customization capability, rather than cost-saving. In a customer service scenario, an 8-second wait for a single conversation will completely ruin the product experience. Since flagship model fine-tuning from top closed-source labs is not open and small closed-source models cannot be deeply customized, small-sized open-source models, through fine-tuning for specific tasks, have become the only choice to support high-frequency real-time interactions.

In the future, enterprise AI will present a division of labor pattern: top closed-source labs will continue to lead the exploration and discovery of new fields, while open-source weight models will increasingly take over the actual production of mature businesses. As model fine-tuning requires high demands on data and talent, the migration from closed-source to open-source will be a slow process lasting several years, during which both will experience continuous growth.