Mini Max M3 Model Official Open Source: Native Multimodular, Million Context

June 16 news.MiniMax It's officially announced yesterdayOpen SourceUnderMultimodalityThe model weight of the flagship model MiniMax M3 and the simultaneous publication of the MSA (MiniMax Sparke Act) technical paper。

Mini Max M3 Model Official Open Source: Native Multimodular, Million Context

Mini Max M3 Total Parameter 428B, Activating Parameter 23B, officially positioned as the first open-source model to introduce multi-modular hybrid training from Step 0。

THE TRAINING PHASE WAS HEAVILY INTEGRATED INTO TEXT, IMAGES AND MULTIMODULAR STAGGERED DATA, WITH A VIEW TO BUILDING A UNIFIED CROSS-MODULAR SEMANTIC SPACE AT THE PRE-TRAINING STAGE; AND THE INTRODUCTION OF THE MSA ARCHITECTURE SIGNIFICANTLY REDUCED THE COSTING UNDER THE LONG CONTEXT SCENARIO。

💻 Github: github.com/MiniMax-AI/MiniMax-M3

Hugging Face: huggingface.co/MiniMaxi/MiniMax-M3

statement:The content of the source of public various media platforms, if the inclusion of the content violates your rights and interests, please contact the mailbox, this site will be the first time to deal with.
Information

Microsoft CEO Nadra proposed "tokeen capital": new coordinates for the competitiveness of businesses in the AI era

2026-6-16 11:15:23

Information

DeepSeek completes $50 billion in finance

2026-6-17 11:12:05

Search