Mistral AI: Mistral Large 3

Overview

Open-weight multimodal Mixture-of-Experts model with 41B active and 675B total parameters, released under Apache 2.0.

Provider

Mistral AI

Modality

Text and image

Input

text, image

Output

text

Weights

Open weights

Supported parameters

tools, structured_outputs, temperatureProvider's model page, checked October 8, 2026.

Distillable

YesLicence or terms, checked October 8, 2026.

Tags

open-weights, multimodal

Docs

Provider documentation

Data policy

Prompts go to

Mistral AI (global endpoint by default)

Retention

Commercial terms: no training on customer data unless opted in, with exceptions, Studio Free mode may train on inputs and outputs (opt-out available), pay-as-you-go can opt out, ZDR available on pay-as-you-go for stateless endpoints

Zero data retention

Available

Region

Global by default, optional EU or US regional inference endpoints at 1.1x pricing

Policy

Mistral AI data policy

Summarised from the provider's published terms. The provider's own policy is what applies.

Where you can run it

How Mistral Large 3 can be deployed, bought and reached through Safeguard.

Deployment

Private cloudOn-premAir-gapped

Procurement

Not published yet

Inference regions

Not published yet

On-prem and air-gapped deployments run wherever you install them.

Pricing

Notes

Standard tier. Cached input $0.05.

Source

Mistral AI pricing , last verified October 8, 2026

No markup on model usage: you pay the provider's list price.

API

OpenAI-compatible. Your API key and access to this model are set up for your workspace when you choose Use with Safeguard.

from openai import OpenAI

client = OpenAI(
    base_url="https://api.safeguard.sh/gpt/v1",
    api_key="SAFEGUARD_API_KEY",
)

response = client.chat.completions.create(
    model="mistral/mistral-large-2512",
    messages=[{"role": "user", "content": "Hello"}],
)
print(response.choices[0].message.content)

Base URL https://api.safeguard.sh/gpt/v1, model mistral/mistral-large-2512.