CatsuCatsu Docs

Mistral AI

Mistral AI embedding provider documentation

Mistral AI offers code-optimized embedding models with quantization support.

Overview

  • Models: 2 models (codestral-embed-2505, mistral-embed)
  • Key Features: Code optimization, binary quantization, batch processing up to 512 texts
  • API Docs: Mistral Embeddings

Environment Variable

export MISTRAL_API_KEY="your-mistral-api-key"

Supported Parameters

ParameterTypeRequiredDescription
modelstrYesModel identifier
inputstr | List[str]YesText(s) to embed (up to 512 texts)
input_typestrNo"query" or "document"
encoding_formatstrNoe.g., "float", "int8"
dimensionsintNoCustom dimensions (codestral-embed-2505 only)
api_keystrNoOverride API key

Examples

Basic Usage

response = client.embed(
    model="mistral-embed",
    input="Hello, Mistral!"
)

Code Embeddings

response = client.embed(
    model="codestral-embed-2505",
    input="def fibonacci(n): return n if n <= 1 else fibonacci(n-1) + fibonacci(n-2)"
)

With Dimensions (codestral-embed-2505)

response = client.embed(
    model="codestral-embed-2505",
    input="Code snippet",
    dimensions=512
)

With Quantization

response = client.embed(
    model="codestral-embed-2505",
    input="Code",
    encoding_format="int8"  # or "binary"
)

Large Batch (up to 512 texts)

# Mistral supports large batches
large_batch = [f"Document {i}" for i in range(500)]

response = client.embed(
    model="mistral-embed",
    input=large_batch
)

print(f"Processed {len(response.embeddings)} texts")

Model Variants

  • codestral-embed-2505 - Code-optimized, 1536d, quantization support
  • mistral-embed - General-purpose, 1024d

For pricing, visit catsu.dev.

Special Notes

  • ✅ codestral-embed-2505 supports dimensions and quantization (float, int8, binary)
  • ✅ Large batch support (up to 512 texts per request)
  • Code-optimized models for software development
  • mistral-embed has fixed 1024 dimensions

Next Steps

On this page