M

Mistral: Devstral 2 2512 (Free)

Mistral AI
Vision
256K
Context
Free
Input /1M
Free
Output /1M
66K
Max Output
Run API Call Open in Chat

Mistral: Devstral 2 2512 (Free)

Description

Devstral 2 is a state-of-the-art open-source model by Mistral AI specializing in agentic coding. It is a 123B-parameter dense transformer model supporting a 256K context window.

The model excels at exploring codebases and orchestrating changes across multiple files while maintaining architecture-level context. It tracks framework dependencies, detects failures, and retries with corrections - solving challenges like bug fixing and modernizing legacy systems.

Devstral 2 can be fine-tuned to prioritize specific languages or optimize for large enterprise codebases. It is available under a modified MIT license.

Technical Specifications

Specification Value
Context Window 262,144 tokens (256K)
Parameters 123 billion (dense transformer)
Input Modalities Text
Output Modalities Text
Max Completion Tokens 65,536
Architecture Dense Transformer

Pricing

Type Price
Input Tokens $0 per 1M tokens (FREE)
Output Tokens $0 per 1M tokens (FREE)
Variant Input Cost Output Cost
mistralai/devstral-2512:free $0/M $0/M
mistralai/devstral-2512 (paid) $0.05/M $0.22/M

Capabilities

Capability Status
Agentic Coding Primary Specialty
Multi-File Orchestration Supported
Codebase Exploration Supported
Framework Dependency Tracking Supported
Bug Detection & Fixing Supported
Legacy System Modernization Supported
Tool/Function Calling Supported
Structured Outputs Supported
JSON Mode Supported
Long-Context Processing Supported (256K)
Fine-Tuning Available
Vision Not Supported

Supported Parameters

Parameter Supported Description
max_tokens Yes Maximum number of tokens to generate
temperature Yes Controls randomness (default: 0.3)
top_p Yes Nucleus sampling parameter
stop Yes Stop sequences for generation
frequency_penalty Yes Reduces repetition of token sequences
presence_penalty Yes Encourages new topics
seed Yes For reproducible outputs
response_format Yes Specify output format (e.g., JSON)
structured_outputs Yes Schema-based structured responses
tools Yes Tool/function calling definitions
tool_choice Yes Control tool selection behavior

Default Parameter Values

Parameter Default Value
temperature 0.3
top_p null
frequency_penalty null

API Usage Example

curl https://api.langmart.ai/v1/chat/completions \
  -H "Authorization: Bearer $LANGMART_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "mistralai/devstral-2512:free",
    "messages": [
      {
        "role": "user",
        "content": "Analyze this Python function and suggest improvements for better error handling."
      }
    ],
    "temperature": 0.3,
    "max_tokens": 4096
  }'

Using with Tools

curl https://api.langmart.ai/v1/chat/completions \
  -H "Authorization: Bearer $LANGMART_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "mistralai/devstral-2512:free",
    "messages": [
      {
        "role": "user",
        "content": "Read the config.py file and update the database connection settings."
      }
    ],
    "tools": [
      {
        "type": "function",
        "function": {
          "name": "read_file",
          "description": "Read contents of a file",
          "parameters": {
            "type": "object",
            "properties": {
              "path": {"type": "string"}
            },
            "required": ["path"]
          }
        }
      },
      {
        "type": "function",
        "function": {
          "name": "write_file",
          "description": "Write contents to a file",
          "parameters": {
            "type": "object",
            "properties": {
              "path": {"type": "string"},
              "content": {"type": "string"}
            },
            "required": ["path", "content"]
          }
        }
      }
    ],
    "tool_choice": "auto"
  }'

Mistral AI Coding Models

  • mistralai/codestral-2508 - Specialized code generation model
  • mistralai/devstral-small-2507 - Smaller development-focused model
  • mistralai/devstral-2512 - Paid version with higher rate limits

Mistral AI General Models

  • mistralai/mistral-large-2512 - Latest large general-purpose model
  • mistralai/mistral-large-2411 - Previous large model version
  • mistralai/mistral-large-2407 - Stable large model
  • mistralai/mistral-medium-3.1 - Mid-tier model
  • mistralai/mistral-small-3.2-24b-instruct-2506 - 24B instruction model
  • mistralai/mistral-small-24b-instruct-2501 - Earlier 24B instruction model
  • mistralai/mistral-nemo - High-traffic general model
  • mistralai/ministral-3b - Lightweight 3B model
  • mistralai/ministral-8b - Mid-size efficient model
  • mistralai/ministral-14b-2512 - 14B parameter model
  • mistralai/mixtral-8x7b-instruct - Mixture of Experts model

Comparable Coding Models (Other Providers)

  • deepseek/deepseek-coder - DeepSeek's coding specialist
  • qwen/qwen-2.5-coder-32b - Qwen's coding model
  • anthropic/claude-3.5-sonnet - Strong coding capabilities
  • openai/gpt-4-turbo - OpenAI's advanced coding support

Model Identification

Field Value
Model Name Devstral 2 2512 (Free)
Model ID mistralai/devstral-2512:free
Version devstral-2-2512
Author/Provider Mistral AI
Created December 9, 2025
License Modified MIT License
Headquarters France

Rate Limits

Limit Type Value
Requests Per Minute (RPM) 600
Requests Per Day (RPD) Unlimited

Provider Information

Mistral AI (Primary Provider)

Field Value
Base URL https://api.mistral.ai/v1
Status Page https://status.mistral.ai/
Data Policy No training use; 30-day retention; no publishing rights
BYOK (Bring Your Own Key) Enabled
Quantization Full precision

Alternative Provider: Chutes (Paid Version Only)

Field Value
Quantization FP8
Max Completion Tokens 65,536
Data Policy Training allowed; retains prompts; non-publishable

Key Use Cases

  1. Agentic Coding - Autonomous code generation and modification with self-correction
  2. Multi-File Refactoring - Orchestrating changes across entire codebases
  3. Bug Fixing - Detecting, diagnosing, and fixing bugs automatically
  4. Legacy System Modernization - Updating old codebases to modern standards
  5. Codebase Exploration - Understanding and navigating large codebases
  6. Framework Integration - Tracking and managing framework dependencies
  7. Enterprise Development - Optimizable for large enterprise codebases

Model Weights

The model weights are publicly available on Hugging Face:

Performance Notes

  • Optimized for agentic coding workflows with autonomous error detection and correction
  • 256K context window enables processing of entire codebases in a single context
  • Tool calling support makes it ideal for IDE integrations and coding assistants
  • Fine-tunable for specific programming languages or enterprise requirements
  • Free tier has same capabilities as paid tier but with rate limits

Data Retention & Privacy

Policy Status
Training Use Disabled (data not used for training)
Prompt Retention 30 days
Publishing Rights Not granted

Source: LangMart API - https://langmart.ai/model-docs:free Last Updated: December 2025