Tech community discussions suggest there is minimal downside to switching to open-weight models, as they are only a few months behind their proprietary counterparts. Additionally, user suspicions regarding silent performance degradation in paid services make open-source and local inference alternatives increasingly appealing for developers.