⌘KCtrl+K
MiniMax M2.7 is a 230B-parameter text-to-text AI model excelling in coding, reasoning, and office tasks.
MiniMax M2.7 is a large language model for complex software engineering, agentic tool use, and office productivity workflows. It is presented as a model deeply participating in its own evolution, with support for complex agent harnesses, dynamic tool search, Agent Teams, and high-fidelity coding and document-editing tasks.
This model is ready for commercial/non-commercial use.
This model is not owned or developed by NVIDIA. This model has been developed and built to a third-party's requirements for this application and use case; see link to Non-NVIDIA MiniMax M2.7 Model Card
GOVERNING TERMS: The trial service is governed by the NVIDIA API Trial Terms of Service; and use of this model is governed by the NVIDIA Open Model License. Additional Information: Modified MIT License. MiniMax M2.7.
Global
Use Case: Designed for advanced coding assistance, agentic workflows, long-horizon software engineering, live production troubleshooting, office document generation and editing, and other complex multi-step productivity tasks.
Build.NVIDIA.com: 04/11/2026 via link
Huggingface: 04/11/2026 via link
References:
Architecture Type: Transformer
Network Architecture: Sparse Mixture-of-Experts (MoE)
Total Parameters: 230B
Active Parameters: 10B
Layers: 62
Hidden Size: 3072
Experts: 256 local experts, with 8 experts activated per token
Input Types: Text
Input Formats: String
Input Parameters: One-Dimensional (1D)
Other Input Properties: Supports long system prompts.
Input Context Length (ISL): 204,800
Output Types: Text
Output Format: String
Output Parameters: One-Dimensional (1D)
Other Output Properties: Not applicable.
Our AI models are designed and/or optimized to run on NVIDIA GPU-accelerated systems. By leveraging NVIDIA's hardware (e.g. GPU cores) and software frameworks (e.g., CUDA libraries), the model achieves faster training and inference times compared to CPU-only solutions.
Runtime Engines:
Supported Hardware:
Operating Systems: Linux
The integration of foundation and fine-tuned models into AI systems requires additional testing using use-case-specific data to ensure safe and effective deployment. Following the V-model methodology, iterative testing and validation at both unit and system levels are essential to mitigate risks, meet technical and functional requirements, and ensure compliance with safety and ethical standards before deployment.
MiniMax M2.7 v2.7
Data Modality: Text
Training Data Collection: Undisclosed
Training Labeling: Undisclosed
Training Properties: Undisclosed
Testing Data Collection: Undisclosed
Testing Labeling: Undisclosed
Testing Properties: Undisclosed
Evaluation Benchmark Score: Publicly reported results include 56.22% on SWE-Pro, 55.6% on VIBE-Pro, 57.0% on Terminal Bench 2, 39.8% on NL2Repo, 1495 ELO on GDPval-AA, 46.3% on Toolathon, 62.7% on MM Claw, and a 66.6% medal rate on MLE Bench Lite. The GDPval-AA result is presented as the highest among open-source models.
Evaluation Data Collection: Undisclosed
Evaluation Labeling: Undisclosed
Evaluation Properties: Evaluation results span software engineering, office productivity, agentic tool-use, and machine learning competition benchmarks, including SWE-Pro, SWE Multilingual, Multi SWE Bench, VIBE-Pro, Terminal Bench 2, NL2Repo, GDPval-AA, Toolathon, MM Claw, and MLE Bench Lite (22 ML competitions). MM Claw testing also cites 97% skill compliance across 40+ complex skills.
| Benchmark | MiniMax M2.7 | Notes |
|---|---|---|
| SWE-Pro | 56.22% | Real-world software engineering benchmark |
| VIBE-Pro | 55.6% | End-to-end project delivery scenarios |
| Terminal Bench 2 | 57.0% | Complex engineering systems understanding |
| NL2Repo | 39.8% | Repository-level engineering benchmark |
| GDPval-AA | 1495 ELO | Professional office domain evaluation; highest among open-source models per launch report |
| Toolathon | 46.3% | Tool-use/generalized interaction benchmark |
| MM Claw | 62.7% | Complex skill and agent workflow evaluation; launch report also cites 97% skill compliance across 40+ skills |
| MLE Bench Lite | 66.6% medal rate | Reported across 22 ML competitions |
| SWE Multilingual | 76.5 | Publicly reported in launch report |
| Multi SWE Bench | 52.7 | Publicly reported in launch report |
Acceleration Engine: vLLM
Test Hardware: NVIDIA H100x4
M2.7 is described as supporting live production debugging workflows involving monitoring metrics, trace analysis, database verification, and SRE-style decision-making. Its use is also described as reducing recovery time for live production incidents to under three minutes on multiple occasions.
M2.7 is positioned as MiniMax's first model deeply participating in its own evolution. During development, the model updated memory, built complex skills for reinforcement learning experiments, improved its learning process based on experiment results, and autonomously optimized a programming scaffold over 100+ rounds for a reported 30% performance improvement.
Recommended deployment settings include temperature=1.0, top_p=0.95, and top_k=40. The model is also described as supporting Agent Teams, dynamic tool search, and multi-agent collaboration in complex agent harnesses.
M2.7 is also described as having strengthened character consistency and emotional intelligence for interactive entertainment use cases. OpenRoom is presented as an interactive demo that places AI interaction in a Web GUI space with real-time visual feedback and scene interactions.
NVIDIA believes Trustworthy AI is a shared responsibility and we have established policies and practices to enable development for a wide array of AI applications. When downloaded or used in accordance with our terms of service, developers should work with their internal model team to ensure this model meets requirements for the relevant industry and use case and addresses unforeseen product misuse.
Please report model quality, risk, security vulnerabilities or NVIDIA AI Concerns here.