Devstral 2 123B Instruct 2512

Description

Devstral 2 123B Instruct 2512 is an agentic large language model designed for software engineering tasks. The model excels at using tools to explore codebases, editing multiple files, and powering software engineering agents, achieving remarkable performance on SWE-bench benchmarks.

This model is ready for commercial/non-commercial use.

Third-Party Community Consideration:

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 Devstral 2 123B Instruct 2512 FP8 Model Card

License and Terms of Use:

GOVERNING TERMS: The trial service is governed by the NVIDIA API Trial Terms of Service. Use of this model is governed by the Modified MIT license.

Deployment Geography:

Global

Use Case:

Use Case: AI code assistants, agentic coding, and software engineering tasks. Designed for developers and organizations leveraging advanced AI capabilities for complex tool integration and deep codebase understanding in coding environments.

Release Date:

Build.NVIDIA.com: 12/2025 via link
Huggingface: 12/2025 via link

Reference(s):

References:

• Mistral AI Model Card
• Mistral Vibe CLI
• Scalable-Softmax Is Superior for Attention
• vLLM Documentation

Model Architecture:

Architecture Type: Transformer
Network Architecture: Mixture-of-Experts with rope-scaling
Total Parameters: 123B

Input:

Input Types: Text
Input Formats: String
Input Parameters: One Dimensional (1D)
Other Input Properties: Text prompts for code generation and agentic coding tasks; supports system prompts for specialized behavior.

Output:

Output Types: Text
Output Format: String
Output Parameters: One Dimensional (1D)
Other Output Properties: Generated code, tool calls, and natural language responses for software engineering tasks.

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.

Software Integration:

Runtime Engines:

• vLLM: >= 0.12.0
• mistral-common: >= 1.8.6

Supported Hardware:

• NVIDIA Ampere: A100
• NVIDIA Blackwell: B200, B100, GB200
• NVIDIA Hopper: H100, H200

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.

Model Version(s)

v1.0 (December 2025) - FP8 quantized variant

Training, Testing, and Evaluation Datasets:

Training Dataset

Data Modality: Text
Training Data Collection: Undisclosed
Training Labeling: Undisclosed
Training Properties: Undisclosed

Testing Dataset

Testing Data Collection: Undisclosed
Testing Labeling: Undisclosed
Testing Properties: Undisclosed

Evaluation Dataset

Evaluation Benchmark Score: 72.2% on SWE-Bench Verified; 61.3% on SWE-Bench Multilingual; 40.5% on Terminal Bench
Evaluation Data Collection: Automated
Evaluation Labeling: Human
Evaluation Properties: SWE-Bench Verified, SWE-Bench Multilingual, and Terminal Bench are standard benchmarks for evaluating software engineering capabilities in AI models.

Inference

Acceleration Engine: vLLM with mistral-common tokenizer
Test Hardware: 8× NVIDIA GPU with tensor parallelism (FP8 precision)

Additional Details

Deployment Options

Devstral 2 can be deployed using:

• Mistral Vibe CLI: Command-line tool for leveraging Devstral capabilities directly in the terminal
• OpenHands: Docker-based deployment for agentic coding workflows
• vLLM Server: Production-ready inference with tensor parallelism support

Recommended Deployment Settings

For optimal performance, deploy with:

• vLLM >= 0.12.0 with --tool-call-parser mistral --enable-auto-tool-choice
• Tensor parallel size of 8 for full model deployment
• Temperature setting of 0.15 for code generation tasks

Compatible Scaffoldings

• Cline
• Kilo Code
• Claude Code
• OpenHands
• SWE Agent

Ethical Considerations

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.