You may be correct but the combination of local models when they are fast enough and work, combined with paying close to zero for deepseek v4 flash from US providers is pretty good. When you need it, glm5.2 is cheap to use and very good for working with larger projects.