Yes, for general language tasks, Google's models win. I wish we had more choices in non-coding space, but even the locally hosted Gemma4 models win against the cheap Sonnet / Luna options.