MiniMax M3: the coding & agentic frontier. A natively multimodal model with MiniMax Sparse Attention (MSA) supporting up to 1M tokens context (512K guaranteed minimum). On Ollama's Cloud it is US-based with zero data retention.
Coding & agentic frontier — autonomous task decomposition, tool invocation, and multi-step reasoning for AI coding assistants and automated workflows.
MiniMax Sparse Attention (MSA) — supports up to 1M tokens context with a guaranteed minimum of 512K, delivering excellent latency and throughput at extreme context lengths.
Native multimodality — multimodal training from step zero with 100T+ pretraining tokens, achieving deep alignment between textual and visual semantic spaces.
Autonomous browsing — scores 83.5 on BrowseComp, surpassing Opus 4.7.
Best practices
Long-range agent tasks — the 1M context is the infrastructure for long-range Agent tasks, long-range Coding, and long-video understanding.
Mixed inputs — combine text and image in the same workflow; multimodality is a native core capability, not an add-on.
Zero data retention — US-based serving on Ollama's Cloud, officially licensed with MiniMax for commercial usage.