MiniMax M3: How new sparse attention cuts AI response time by 15.6X
MiniMax’s upcoming M3 model introduces a custom sparse attention mechanism that accelerates long-context AI responses by 15.6 times without sacrificing reasoning accuracy, promising a breakthrough for enterprise agent deployments and multimodal workflows.