MiniMaxAI/MiniMax-M3
Summary
MiniMax releases M3, a native multimodal model with 1M context and ~428B parameters, using MiniMax Sparse Attention (MSA) for efficient long-context processing, achieving frontier-level coding and agentic performance.
View Cached Full Text
Cached at: 06/12/26, 02:52 PM
MiniMaxAI/MiniMax-M3 · Hugging Face
Source: https://huggingface.co/MiniMaxAI/MiniMax-M3
MiniMax-M3 is a native multimodal model with 1M context. It has ~428B parameters and ~23B activated parameters.
Highlights:
- **Native Multimodality:**M3 undergoes mixed-modality training from the very first step, enabling deeper semantic fusion across text, image, and video.
- **Context Scaling via Sparse Attention:**M3 introduces MiniMax Sparse Attention (MSA) to improve long context efficiency. M3 delivers 9× prefill and 15× decode speedups compared to M2 at 1M context, reducing per-token compute to 1/20.
- **Coding & Cowork Capability:**M3 achieves frontier-level performance across long-horizon agentic benchmarks, excelling in both coding and cowork.
https://huggingface.co/MiniMaxAI/MiniMax-M3#minimax-sparse-attention-msaMiniMax Sparse Attention (MSA)
M3 is powered byMiniMax Sparse Attention (MSA), a high-performance sparse attention operator designed for million-token contexts. Compared with GQA, MSA dramatically reduces the attention compute and memory footprint while preserving model quality.
📄 Read the technical report:arXiv:2606.13392·Hugging Face Papers
https://huggingface.co/MiniMaxAI/MiniMax-M3#how-to-useHow to Use
M3 supports two reasoning modes:
- thinking— for complex reasoning, agentic tasks, and long-horizon collaboration.
- non-thinking— for latency-sensitive scenarios such as chat and code completion.
https://huggingface.co/MiniMaxAI/MiniMax-M3#local-deploymentLocal Deployment
Download the model:
hf download MiniMaxAI/MiniMax-M3 --local-dir MiniMax-M3
We recommend the following inference frameworks (listed alphabetically) to serve the model:
- SGLang- seeSGLang cookbook.
- vLLM- seevLLM recipes.
- Transformers- seeTransformers docs.
https://huggingface.co/MiniMaxAI/MiniMax-M3#inference-parametersInference Parameters
We recommend the following parameters for best performance:temperature=1\.0,top\_p=0\.95,top\_k=40.
https://huggingface.co/MiniMaxAI/MiniMax-M3#contact-usContact Us
Contact us at[email protected].
Similar Articles
MiniMax M3 (2 minute read)
MiniMax introduces M3, the first open-weights model to combine coding, agentic, and multimodal capabilities with up to 1M context via sparse attention.
MiniMax teases upcoming M3 model with new sparse attention mechanism and 15.6X long-context response speed boost (12 minute read)
MiniMax has released a detailed technical report on its M2 series and teased the upcoming M3 model, which uses a novel sparse attention mechanism to achieve up to 15.6× faster decoding at million-token contexts.
MiniMax promises M3 weights after 1M-context model launch (2 minute read)
MiniMax released M3, a model with a 1M-token context window and native multimodal input, via API. The company promises open-weight release and a technical report within 10 days.
MiniMaxAI/MiniMax-M2.7
MiniMaxAI releases MiniMax-M2.7, an open-weight model featuring self-evolution capabilities, advanced agent team support, and strong performance on software engineering benchmarks (56.22% on SWE-Pro, 66.6% medal rate on MLE Bench Lite), with notable applications in production incident recovery and professional work tasks.
MiniMax Sparse Attention
MiniMax Sparse Attention introduces a blockwise sparse attention mechanism that achieves significant speedups for ultra-long-context LLMs, reducing per-token attention compute by 28.4x at 1M context with wall-clock speedups of 14.2x for prefill and 7.6x for decoding on H800 GPUs. The method is accompanied by an open-source inference kernel and a publicly released multimodal model.