MiniMax M3.1 Flash releases open weights alongside independent comparisons

MiniMax M3.1 Flash weights are downloadable: a 428B-total/23B-active multimodal MoE with million-token context and SGLang/vLLM deployment guidance. Earlier Artificial Analysis comparisons reported reasoning gains and roughly halved evaluation cost while flagging factual-answering weakness; the latest release coverage adds no benchmarks.