Mistral Large

Mistral Large is Mistral AI's flagship large language model, offering strong performance with a focus on efficiency and multilingual capability.

Key Specifications

• Context window: 128K tokens
• Max output tokens: 8,192
• Architecture: Transformer
• Modalities: Text (plus Pixtral for vision)
• Pricing: $2.00 / $6.00 per 1M input/output tokens
• Availability: La Plateforme, cloud partners

Capabilities

• Multilingual — Strong in French, German, Spanish, Italian, and 10+ languages
• Code generation — Good coding performance
• Function calling — Native tool use
• JSON mode — structured-output support
• Agentic — Designed for agent workflows

Ecosystem

Mistral offers multiple models:

• Mistral Large — Flagship (this page)
• Mistral Small — Efficient, fast
• Pixtral Large — Multimodal variant
• Codestral — Code-specialized
• Les Ministraux — Edge-device optimized (0.3B-1.5B)

Architecture Details

Uses GQA with RoPE positional encoding. Trained with pretraining on diverse multilingual data including web, code, and scientific content.

Comparison

• GPT-4o — Competitor in pricing tier
• Llama 4 — Open-source competitor
• qwen-2.5 — Multilingual competitor
• See LLM API Pricing Comparison for cost comparison
• See Open-Source vs Closed-Source LLMs for openness comparison