36Kr's feature on Zhixiang Future (智象未来) argues that multimodal world models have a token profit margin far exceeding that of language models. The argument: each multimodal token (e.g., a video frame) carries more information and higher value than a text token, justifying higher pricing.