MiniMax has launched M1, an open-weight hybrid attention model built for long-context reasoning and complex tasks. With 456 billion parameters and support for one million-token inputs, MiniMax-M1 ...