2 ms·Text or pixels? On the token efficiency of visual text inputs in multimodal LLMs2 points by hhs 1y ago