Mistral Large 4
AI model · Multimodal model · Text → Text by Mistral AI
State-of-the-art open-weight multimodal model for agentic coding and cybersecurity.
Mistral Large 4 is Mistral AI's state-of-the-art, open-weight, general-purpose multimodal model designed for complex reasoning and agentic tasks. It features a granular Mixture-of-Experts architecture with 52B active parameters and 1.05T total parameters, complemented by a 1.6B vision encoder. The model is built for high-consequence environments, excelling in agentic coding, cybersecurity, and long-context document analysis. It supports a 1M token context window and is available through Mistral's API with regional data processing controls or as open weights for self-hosted deployments.
- Languages
- English, French, German, Spanish, Italian, Portuguese, Chinese, Japanese, Korean
- Active parameters
- 52B
- Total parameters
- 1.05T
- Vision encoder parameters
- 1.6B
- Context window
- 1M tokens
- Modalities
- Text, Image → Text
Overview
Mistral Large 4 is Mistral AI's state-of-the-art, open-weight, general-purpose multimodal model designed for complex reasoning and agentic tasks. It features a granular Mixture-of-Experts architecture with 52B active parameters and 1.05T total parameters, complemented by a 1.6B vision encoder. The model is built for high-consequence environments, excelling in agentic coding, cybersecurity, and long-context document analysis. It supports a 1M token context window and is available through Mistral's API with regional data processing controls or as open weights for self-hosted deployments.
Architecture and reasoning
- Mixture-of-Experts architecture: Features 52B active parameters and 1.05T total parameters in a granular Mixture-of-Experts architecture.
- Native vision capabilities: Includes a 1.6B vision encoder for processing multimodal inputs.
- Agentic reasoning: Optimized for complex agentic workflows, multi-step reasoning, and cybersecurity tasks.
- Developer features: Supports structured outputs, function calling, and document Q&A.
Deployment and context
- 1M token context window: Supports a context window of up to 1M tokens for long-document analysis.
- Open-weight availability: Available as open-weights for independent operation and evolution.
- Regional API access: Accessible via the Mistral API with regional inference controls in the EU and US.
Pricing
- Input per 1M tokens
- $0.7
- Cached input per 1M tokens
- $0.07
- Output per 1M tokens
- $2.1
Specifications
- Active parameters
- 52B
- Total parameters
- 1.05T
- Vision encoder parameters
- 1.6B
- Context window
- 1M tokens
- Modalities
- Text, Image → Text
Added by AntSpace from public sources (docs.mistral.ai).