MINIMAX 4M CONTEXT LENGTH MODEL

Summary from the TLDR AI newsletter

The next frontier level model from China uses a novel lightning attention. It is the first linear variant that achieves competitive performance with some of the best frontier models. The model has 400B+ parameters and was trained with 4 million tokens in context. The researchers have released a strong technical report, model weights, and a code repository. There is also a companion vision model.