Last updated 2026-05-17

Mistral Provider

The mistral provider implements Provider (chat) and Embedder against api.mistral.ai. Mistral's REST surface is OpenAI-compatible, so this provider is a thin wrapper that pins the base URL and known model ids; under the hood it reuses the OpenAI provider machinery.

Import path

import (
    "github.com/elloloop/llmrouter"
    "github.com/elloloop/llmrouter/providers/mistral"
)

Construction

p, err := mistral.New(
    llmrouter.WithAPIKey(os.Getenv("MISTRAL_API_KEY")),
)

Default endpoint: https://api.mistral.ai/v1. Auth via Authorization: Bearer <key>.

Chat models

mistral-large-latest — flagship.
mistral-small-latest — balanced cost/quality.
codestral-latest — code-specialised.
open-mistral-nemo — open-weights 12B.
pixtral-large-latest — multimodal (image input).
ministral-8b-latest, ministral-3b-latest — small models.

Chat example

stream, err := p.CompletionStream(ctx, llmrouter.ChatRequest{
    Model: "mistral-large-latest",
    Messages: []llmrouter.Message{
        llmrouter.TextMessage("system", "You are concise."),
        llmrouter.TextMessage("user", "What is Mixture of Experts?"),
    },
})

for chunk := range stream.Chunks() {
    for _, c := range chunk.Choices {
        fmt.Print(c.Delta.Content)
    }
}

Embeddings

Mistral exposes a single embedding model:

mistral-embed — 1024 dims, multilingual.

resp, err := p.Embed(ctx, llmrouter.EmbedRequest{
    Model:  "mistral-embed",
    Inputs: []string{"Bonjour le monde"},
})

TaskType is not modelled on the wire; the library ignores it for Mistral.

Error handling

Non-2xx responses surface as *llmrouter.ErrUpstream with Provider == "mistral".

Caveats

Strict JSON mode. Mistral's response_format is OpenAI-compatible; pass via Raw.
EU data residency. Mistral hosts in the EU by default. For US users, latency is slightly higher than US-hosted vendors.