← Back to all models

Mistral Small

Mistral AILarge Language ModelOpen Source

Mistral's cost-efficient open-weight model optimized for simple tasks and high-throughput applications. Apache 2.0 license.

Abilities

Text GenerationCode GenerationSummarizationMultilingualOn-device

Use Cases

ChatbotAutomationOn-device

Available in Tools

Availability

How to Use

Use via Mistral API or download from Hugging Face. Can run locally with Ollama or vLLM.

Pros

  • Fully open with Apache 2.0 license — unrestricted commercial use
  • Extremely cost-efficient at $0.10/1M input — excellent for batch processing
  • Fast response times — suitable for real-time applications
  • Can run on a single consumer GPU (24GB VRAM) with quantization
  • Good multilingual support, especially European languages

Cons

  • Significantly less capable than Large on complex reasoning and analysis
  • 32K context window — much shorter than competitors
  • Not suitable for hard coding or complex math tasks
  • Limited multimodal capabilities — text only

What to Use It For

star

Perfect For

Budget-friendly batch processing

At $0.10/1M input tokens it is one of the cheapest open model APIs — ideal for high-volume simple tasks

Simple classification tasks

Fast and cheap enough for real-time text classification without overkill model capabilities

Local deployment on single GPU

24B parameters fit on a single consumer GPU (24GB VRAM) with quantization — easy self-hosting

thumb_up

Good For

Summarization and FAQ bots

Good enough quality for straightforward summarization and structured Q&A at minimal cost

Text extraction

Reliable at pulling structured data from documents without needing frontier model intelligence

warning

Not Recommended

Complex reasoning or analysis

24B model lacks the depth for multi-step reasoning and nuanced analytical tasks

Try instead: Mistral Large, Claude Sonnet 4

Long document processing

32K context window is too short for processing large documents or codebases

Try instead: Gemini 2.5 Pro

block

Do Not Use For

Hard coding tasks

Produces too many bugs on complex code — not reliable enough for serious development work

Try instead: Codestral, Claude Sonnet 4

Multimodal tasks

Text-only model with no vision, audio, or image capabilities at all

Try instead: GPT-4o

Technical Details

Pricing$0.10 / 1M input, $0.30 / 1M output
Parameters24B
detail.contextWindow32K tokens
LicenseApache 2.0

Links