Which model should I use for my project?

For general tasks/coding: Mistral Medium. For documents: OCR. For audio: Voxtral. For cost-sensitive work: Mistral Small. See the model overview for specs.

How is API pricing calculated?

Most models are priced per million tokens (input/output separately). Batch processing cuts cost by 50%. Cached input tokens reduce cost by up to 90%. Some APIs (OCR, speech) use per-page or per-minute pricing.

What's the difference between Free and Pro plans?

Pro offers 6x more messages, 5x more web searches, unlimited task scheduling, all-day coding, and email support. Free is for light testing.

Can I self-host Mistral models?

Yes. All models can be self-hosted. Open-weight models (e.g., Mistral 7B) use Apache 2.0 for research. Commercial deployments require a separate Mistral license.

Is there a student discount?

Verified students can get the Pro plan for $5.99/month (normally $14.99).

What's the core model architecture?

Flagship models (Medium, Large) use a Mixture of Experts (MoE) architecture for efficient, high-performance inference across many language tasks.

Directory/Coding/AI Assistants/Mistral AI

Mistral AI

Mistral AI is a comprehensive platform for building, training, and deploying AI solutions, offering products like Vibe (an AI agent for long-horizon tasks), Studio (for AI app development), Forge (for custom model training), and Compute (for scalable infrastructure). It provides open-weight models such as Mistral 7B and services focused on AI agent orchestration, model customization, and enterprise-grade security, with a strong emphasis on sovereignty and privacy. Targeted at developers, founders, and enterprises, Mistral AI enables the creation of tailored AI systems with tools for coding, document intelligence, and speech, supported by a freemium pricing model that includes free and paid plans.

Coding

Visit website

Added on

Sep 16, 2026

Monthly Visits

10.7M

Vibe AI agentStudio app builderForge model trainingAI Cloud infrastructureMistral MediumMistral SmallOCRVoxtral TTSCoding agentsImage generation

Product information

Everything worth knowing about Mistral AI

The complete picture — from everyday features to the technical detail builders and IT folks dig for.

Core features

What it actually does

Vibe AI agent

AI agent for long-horizon tasks, fluent in knowledge and tools.

Studio

Build, test, and run AI agents and apps with enterprise-grade tooling.

Forge

Train, align, and evaluate custom AI models.

Vibe for Code

Coding agents in the terminal, IDE, and background for faster shipping.

Image Generation

State-of-the-art image creation capabilities.

Technical capabilities

Agent Orchestration

End-to-end orchestration and observability for AI applications.

Dedicated GPU Clusters

Frontier-grade infrastructure for training and inference.

Must watch videos

Generate playable HTML5 games with Mistral Medium and GLM

8:13

Tutorial8:13

Generate playable HTML5 games with Mistral Medium and GLM

Learn how to use agents to generate complete, playable HTML5 games from text descriptions with automated quality review and iterative fixes. Follow along with the code and docs: Follow along → https://mistr.al/4xbU35T Get your API key → https://mistr.al/3UCqVH5

Build a database advisor agent with a custom DeepWiki Connector

6:36

Tutorial6:36

Build a database advisor agent with a custom DeepWiki Connector

In this tutorial, we build an AI agent that reads the actual source code of three popular databases: SQLite, DuckDB, and LevelDB, and recommends the best one for a specific workload. We walk through the complete Connectors lifecycle: create, list, use, update, and delete while evaluating the databases using an Agent. Follow along with the notebook → https://mistr.al/4xgGL8j Connectors docs → https://mistr.al/45Gg7Kl Mistral Studio → https://mistr.al/4fEYH6J

Mistral and Microsoft Expand Global Strategic Partnership to Give Enterprises AI They Can Control

5:01

Tutorial5:01

Mistral and Microsoft Expand Global Strategic Partnership to Give Enterprises AI They Can Control

Mistral Co-Founder & CEO Arthur Mensch and Microsoft President & Vice-Chair Brad Smith discuss the outlook of the two companies’ expanded strategic partnership. The partnership spans compute, the integration of Mistral models into Microsoft products and a joint go-to-market.

Who uses it

Real use cases, no hype

Autonomous Coding & DevOps

Generate, refactor, and review code via CLI/IDE. Automate CI/CD and legacy translation.

Document Intelligence & OCR

Extract structured data from complex PDFs, images, and documents at scale (Mistral OCR).

Custom Model Fine-tuning

Train domain-specialized models on proprietary data using Forge for alignment and evaluation.

Long-Horizon AI Agents

Deploy persistent agents (Vibe) for multi-step research, analysis, and task execution.

Enterprise Knowledge Search

Connect internal data sources to build a secure, queryable company knowledge base.

AI Application Development

Build, orchestrate, and deploy production AI agents/apps with full control using Studio.

Speech & Voice Processing

Integrate speech-to-text and text-to-speech (Voxtral) into applications for voice interfaces.

What's great

Open-weight models (Apache 2.0 license for research)
Multi-platform: Web, CLI, IDE, and mobile apps
Strong EU data hosting and sovereignty focus
Full-stack platform: from agents (Vibe) to training (Forge)

Technical strengths

Mixture of Experts (MoE) architecture in flagship models
Efficient token use with caching (up to 90% cost reduction)
Self-hosting option for all models
Enterprise-grade tooling for observability and guardrails

Where it falls short

Free tier is heavily limited (messages, search, coding)
Can feel enterprise-first; DIY dev experience can lag
Pricing for high-volume usage is opaque without a custom plan

Technical limitations

Model selection (Small, Medium, Large) requires careful evaluation
Proprietary tooling (Studio, Forge) may create vendor lock-in
API pricing structure varies per model/service (per token vs per page)

For technical folks

The deep-dive specs

Core Model Architecture

Proprietary transformer-based models (e.g., Mistral Large, Mistral Medium) alongside open-weight models (Apache 2.0).

Deployment Options

SaaS (Mistral Cloud EU), Self-hosted (on-prem/virtual cloud), Partner Clouds (AWS, GCP, Azure, SAP, IBM, Snowflake, NVIDIA, Outscale).

Pricing Model (API)

Per million tokens (input/output). Example: Mistral Large $2/M in, $6/M out. Batch processing 50% discount.

Primary AI Modalities

Text, Vision (multimodal), Speech (Voxtral TTS & transcription), OCR, Code generation.

FAQ

Includes technical Q&A

Yes. The Free plan includes basic Vibe chat, web search, image generation, and limited coding sessions. You also get $10 in monthly API credits for Studio.