MiniMax Releases M3, an Open-Weight Multimodal and Long-Context Model
MiniMax Research has officially released M3, an open-weight model that combines frontier-level coding and agentic performance with a 1M token context window and native multimodality. M3 utilizes a new attention architecture called MSA (MiniMax Sparse Attention) and excels on benchmarks like SWE-Bench Pro and OmniDocBench, surpassing several other leading models. It supports image and video input and can autonomously operate a desktop computer.
M3 is the first open-weight model to integrate frontier coding, 1M context, and native multimodality, making advanced AI capabilities more accessible to developers. Its strong performance in agentic tasks and ability to operate computers autonomously could drive innovation in AI automation.
Learn one new AI thing every day.
Daily Deck sends you seven plain-English cards like this every morning. Free.
Start free