@seclink: MiniMax M3 is now open source. Generally, those willing to open source are outdated, with undisclosed firepower still unreleased. Hugging Face main repository (recommended): https://huggingface.co/MiniMaxAI/MiniMax-M3… Here provides the complete model weights...
Summary
MiniMax has open-sourced its native multimodal large model M3, with approximately 428B total parameters (~23B active), supporting 1M context length, and introducing MiniMax Sparse Attention (MSA) technology to improve long-context efficiency.
View Cached Full Text
Cached at: 07/14/26, 10:22 AM
MiniMax M3 has been open-sourced.
Generally, only outdated models are open-sourced; the latest ammunition remains undisclosed.
Hugging Face Main Repository (Recommended):
https://huggingface.co/MiniMaxAI/MiniMax-M3… Here you can find the full model weights (~428B total parameters, ~23B activated parameters, native multimodal, supports 1M context), available in Safetensors/PyTorch format, along with quantized versions (e.g., MiniMaxAI/MiniMax-M3-MXFP8).
GitHub Repository (Supporting Code and Documentation): https://github.com/MiniMax-AI/MiniMax-M3… Includes inference guides, MSA (MiniMax Sparse Attention) support, and more.
MiniMaxAI/MiniMax-M3 · Hugging Face
Source: https://huggingface.co/MiniMaxAI/MiniMax-M3 MiniMax
MiniMax Agent (https://agent.minimax.io/)API (https://platform.minimax.io/docs/guides/text-generation)MiniMax Website (https://www.minimax.io/) ModelScope MiniMax AI (https://modelscope.cn/organization/minimax)WeChat (https://platform.minimaxi.com/docs/faq/contact-us)Discord (https://discord.com/invite/DPC4AHFCBw)Hugging Face (https://huggingface.co/MiniMaxAI)GitHub (https://github.com/MiniMax-AI/MiniMax-M3)arXiv Paper (https://arxiv.org/abs/2606.13392)LICENSE (https://huggingface.co/MiniMaxAI/MiniMax-M3/blob/main/LICENSE)
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) (https://github.com/MiniMax-AI/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.
GQA vs MSA Efficiency Comparison
📄 Read the technical report:arXiv:2606.13392 (https://arxiv.org/abs/2606.13392)·Hugging Face Papers (https://huggingface.co/papers/2606.13392)
https://huggingface.co/MiniMaxAI/MiniMax-M3#how-to-useHow to Use
- MiniMax Agent (https://agent.minimax.io/)
- MiniMax API (https://platform.minimax.io/)
M3 supports three reasoning modes through thethinkingparameter:
enabled— Reasoning is always enabled.adaptive— M3 automatically determines when additional reasoning is beneficial.disabled— Reasoning is disabled to minimize latency and maximize throughput.
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 to serve the model:
- SGLang (https://docs.sglang.io/)- seeSGLang cookbook (https://docs.sglang.io/cookbook/autoregressive/MiniMax/MiniMax-M3).
- vLLM (https://github.com/vllm-project/vllm)- seevLLM recipes (https://recipes.vllm.ai/MiniMaxAI/MiniMax-M3).
- Transformers (https://github.com/huggingface/transformers)- seeTransformers docs (https://huggingface.co/docs/transformers/model_doc/minimax_m3_vl).
- KTransformers (https://github.com/kvcache-ai/ktransformers)- seeKTransformers MiniMax-M3 tutorial (https://github.com/kvcache-ai/ktransformers/blob/main/doc/en/kt-kernel/MiniMax-M3-Tutorial.md).
- unsloth (https://unsloth.ai/)- seetutorial (https://unsloth.ai/docs/models/minimax-m3)
https://huggingface.co/MiniMaxAI/MiniMax-M3#inference-parametersInference Parameters
We recommend the following parameters for best performance:temperature=1\.0,top\_p=0\.95.
https://huggingface.co/MiniMaxAI/MiniMax-M3#contact-usContact Us
Contact us at[email protected].
Similar Articles
MiniMaxAI/MiniMax-M3
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.
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.
@TeksEdge: With MiniMax M3 open source now out, here is what to expect on quants and sizes, including VRAM needed: MiniMax M3 (428…
MiniMax M3, a 428B MoE model with ~23B active parameters, is now open source. It offers ultra-long context (up to 1M) and efficiency improvements, with various quantized sizes and VRAM requirements for local deployment.
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.
@PrajwalTomar_: Okay this is wild. MiniMax just dropped M3, and it might be the most capable open model for building right now. I gave …
MiniMax released M3, an open-source AI model that leads coding benchmarks and offers a 1M token context for handling entire codebases.