M

Mistral: Devstral 2 2512

Mistral AI
262K
Context
$0.0500
Input /1M
$0.2200
Output /1M
N/A
Max Output
Run API Call Open in Chat

Mistral: Devstral 2 2512

Model ID: mistralai/devstral-2512 Provider: Mistral AI Category: Code-Specific, Coding Agent Release Date: December 9, 2025 Parameters: ~123 billion License: Modified MIT

Overview

Devstral 2 2512 is a state-of-the-art open-source model specializing in agentic coding. With 123B parameters and 256K context window support, it excels at exploring codebases while preserving architectural understanding and implementing changes across multiple files. The system includes failure detection with automatic correction capabilities and can tackle challenges like legacy system modernization and bug fixes.

Technical Specifications

Property Value
Parameters ~123 billion
Context Length 262,144 tokens
Architecture Dense Transformer
Input Modalities Text
Output Modalities Text
Max Completion Tokens 65,536

Pricing

Standard Variant (via Chutes)

Type Price
Input $0.05 per 1M tokens
Output $0.22 per 1M tokens

Alternative: Free Tier Available (via Mistral)

Mistral AI offers free tier access through their official API

Capabilities

Coding Specialization

  • Codebase exploration with architectural awareness
  • Multi-file change implementation
  • Failure detection with automatic correction
  • Legacy system modernization
  • Bug fix and debugging
  • Code refactoring and optimization
  • Architecture-preserving modifications

Input Modalities

  • Text only

Output Modalities

  • Text only

Key Features

  • 256K (262,144) token context window
  • Dense transformer architecture
  • Failure detection and correction
  • Multifile code changes
  • Structured outputs support
  • Tool calling capabilities

Supported Parameters

  • temperature - Sampling temperature control
  • top_p - Nucleus sampling parameter
  • stop - Stop sequences for output termination
  • frequency_penalty - Reduce repetitive tokens
  • presence_penalty - Encourage diverse topics
  • seed - Random seed for reproducibility
  • top_k - Top-K sampling parameter
  • repetition_penalty - Control repetition
  • response_format - Control output format
  • tools - Available tools/functions
  • tool_choice - Tool selection strategy
  • structured_outputs - Enable structured outputs

Use Cases

  • Code Analysis: Large codebase understanding and analysis
  • Development: Multi-file modifications and refactoring
  • Debugging: Bug detection and fixing across files
  • Architecture: System architecture understanding and preservation
  • Modernization: Legacy code modernization and updates
  • Agentic Workflows: Autonomous coding agent capabilities
  • Mistral: Mistral Large 2
  • Mistral: Mistral Medium
  • Mistral: Mistral Small
  • DeepSeek: Devstral alternatives for coding

Providers

Primary Provider: Chutes

Alternative Provider: Mistral

Performance Metrics

Usage Statistics (December 12-15, 2025)

  • Peak activity: Over 300,000 requests
  • Prompt tokens processed: 1.4B+ on highest-volume day
  • Sustained high usage indicating strong developer adoption

Model Access

  • Chat Interface: Available
  • Model Comparison: Enabled
  • Weights: Available on HuggingFace (mistralai/Devstral-2-123B-Instruct-2512)

Data & Privacy

  • Training Use: Varies by provider
  • Prompt Retention: Provider-dependent
  • Publishing: Check provider-specific policies

Additional Notes

  • 123B parameter count makes it suitable for complex coding tasks
  • 256K context window enables analysis of large codebases
  • Free tier availability through Mistral makes it accessible to developers
  • Excellent choice for autonomous coding agents and multi-file changes
  • Architecture-aware modifications preserve system design