AI model by Mistral AI
Mistral Large 3 for business
Mistral's open-weight general-purpose flagship under Apache 2.0, with image input, tool calling, structured outputs and a 256K token window.
- General purpose
- Open weights
- Vision and multimodal
- Open weights
- Reviewed on September 24, 2026
- Mistral AI
Capability tiersRelative, not benchmarks
Key facts about Mistral Large 3
- Model ID at review
- mistral-large-2512
- Provider
- Mistral AI
- Main category
- General purpose
- Open weights
- Yes, Apache 2.0
- Last reviewed
- September 24, 2026
Fit
Where it fits and where it does not
Good at
-
General assistant, drafting and extraction work.
-
Image input with text.
-
Tool calling and structured outputs.
-
A permissive Apache 2.0 license for self-hosting.
-
Use through the Mistral API or on your own hardware.
Not the right choice for
-
Small servers, since it is a very large model.
-
Teams that only need simple, high-volume tasks.
-
Audio or video input.
-
Decisions without a test on your own data.
Use cases
Business use cases we would use it for
Controlled hosting
Our notes
When we would choose it
Mistral Large 3 was released in December 2025. Mistral describes it as an open-weight, general-purpose multimodal model, and publishes it under the Apache 2.0 license. It accepts images, supports tool calling and structured outputs, and has a 256K token context window. It is a mixture-of-experts model with 675 billion parameters in total and 41 billion active per token.
The combination of a permissive license and strong capability is unusual. It means a business can use the same model through the Mistral API today and move it to its own infrastructure later, without changing models. For companies with strict data rules, that flexibility is valuable.
Self-hosting a model this size is a serious project, though. For most teams, the Mistral API or a cloud provider is the practical way to use it. If you need something smaller to host yourself, look at Mistral Small 4.
We would include Large 3 in a comparison with Claude Sonnet, GPT-6 Sol and DeepSeek, especially where cost, European hosting or open weights matter. See AI model evaluation.
Mistral also lists Mistral Medium 3.5, released in April 2026 as a newer open-weight model for agent and coding work under a modified MIT license. It is worth adding to the same test if your work leans toward agents or code.
Before you commit
Things to check before you commit
-
Hardware for self-hosting
It is a large mixture-of-experts model. Price the hardware before planning to host it.
-
Data terms
If you use the Mistral API, check data terms and regions.
-
Quality
Compare with hosted leaders on your own tasks.
-
Price level
The API price is low for its class. Confirm current rates.
Alternatives
Models to compare it with
Mistral Small 4
Mistral's efficient open-weight model under Apache 2.0, one hybrid model for instructions, reasoning and coding, with tool calling and a 256K window.
Llama 4 Maverick
The larger Llama 4 open-weight model, with image input and a 1 million token context window, for self-hosted assistants and analysis.
DeepSeek V4.1 Flash
DeepSeek's default, best-value model: open weights under MIT, image input, tool calling, JSON output, optional thinking and a 1 million token window.
Keep exploring
Solutions, services and guides
Related services
View all related services- LLM integration Add a large language model to software you already have, with the guardrails, costs and logging handled.
- AI chatbots Website and messaging chatbots that answer common questions well and hand everything else to a person.
- AI Product Development AI agents, knowledge assistants, copilots and document automation built into the way your team already works.
- Document automation Read invoices, forms, contracts and IDs, pull out the right fields and route them for review.
- AI copilots AI helpers built into your product or internal tools that draft, summarize and suggest while people decide.
- AI workflows Step-by-step automations where AI handles the reading, sorting and drafting inside a process you control.
Solutions
View all solutions- AI document processing Read forms, applications, IDs and statements, extract the fields you need and route each document for the right review.
- Knowledge assistant (RAG) Ask a question in plain words and get an answer from your own documents, with links to the sources.
- AI-assisted content workflows Briefs, first drafts, edits and repurposing in a workflow where people set the angle and approve every word.
- Feedback analysis Read every review, survey and ticket, group them by theme and sentiment, and show what customers keep asking for.
- Screening assistant Summarize each application against your criteria, flag strong matches and missing information, and leave every decision to a person.
- AI translation workflows Translate websites, products and support content quickly with AI, a shared glossary and native-speaker review where it matters.
Industries
View all industries- Finance and accounting Client portals, document collection, invoice and receipt processing, and reporting for accounting firms and finance teams.
- Legal Contract review assistants, knowledge search, intake and document automation for law firms and in-house legal teams.
- Education and elearning Course platforms, student portals, content workflows and AI tutoring assistants for schools and training companies.
Case studies
View all case studiesGuides and articles
View all guides and articles- How to choose an AI model for your business A step-by-step way to pick an AI model by testing candidates on your own tasks, data rules and budget.
- What is RAG and when do you need it? How retrieval-augmented generation lets AI answer from your documents, when it fits and how to build it well.
- How to keep customer data safe when using AI Practical steps to protect customer data when you use AI services, from data terms to access and logging.
- RAG vs fine-tuning Two ways to make AI work with your knowledge, compared by cost, accuracy and upkeep.
MCP servers
View all mcp servers- Files and storage Servers that connect AI to cloud drives and object storage, so it can find, read and organize documents.
- AWS S3 Amazon S3 is handled through the managed AWS MCP Server, which lets AI run scripted AWS actions, create presigned links and search AWS docs.
- Box Box's official MCP server lets AI search and read files, ask questions across documents with Box AI, extract data and upload files.
- Dropbox Dropbox's own MCP server lets AI list, search and read files, convert them to markdown and create, move, share and delete files. In beta.
- Google Drive Google's own Drive MCP server lets AI search, read and create files in Google Drive, with your Google Workspace sign-in.
- MySQL A community MCP server that lets AI run SQL against MySQL, read-only by default, with insert, update and delete each switched on separately.
Glossary terms
View all glossary terms- Large language model A large language model, or LLM, is an AI model trained on vast amounts of text that can understand and generate language, and often images and code.
- Structured output Structured output is when an AI model returns its answer in a fixed format, such as JSON matching a schema, so software can use it reliably.
- LLM integration LLM integration is connecting a large language model to your software and data, so AI features work inside your own products and processes.
- Open-weights model An open-weights model is an AI model whose trained parameters are published, so anyone can download and run it on their own hardware under its license.
- Artificial intelligence Artificial intelligence is the broad field of building software that performs tasks that normally need human judgment, such as understanding language or images.
- Context window A context window is the maximum amount of text, measured in tokens, that an AI model can consider at once, including the question, documents and its answer.
FAQ
Questions people ask us
Have a question that is not here? Ask us directly.
It is released with open weights under the Apache 2.0 license.
Yes. Mistral lists image input, tool calling and structured outputs.
256K tokens, according to the model card.