I think the practical economics of the LLM business are becoming clearer in recent times. Huge models are expensive to train and expensive to run. As long as it meets the average user's everyday needs, it's probably much more profitable to just continue with multimodal and fine-tuning development on smaller models.