Mistral Large
Mistral AI's flagship multilingual text model, released in 2024 and now deprecated.
Mistral Large is a proprietary text-generation model from the French lab Mistral AI, launched in 2024 as its flagship offering. It handles reasoning, translation and coding across several European languages, and was later retired in favour of newer models.
- Flagship Mistral AI text model released in 2024, now deprecated
- 32K-token context window for recall from long documents
- Native function calling plus a constrained output mode
- JSON format mode emits valid JSON for structured pipelines
- First Mistral model distributed through Microsoft Azure
What it is
Mistral Large is a text-generating language model put out by Mistral AI in 2024. The company presented it as its top model at the time, aimed at hard reasoning work, rewriting and understanding of text, and writing code. It works in English plus French, Spanish, German and Italian, with attention to grammar and cultural detail. A 32,000-token window lets it pull exact details out of long documents. Mistral AI has since stopped maintaining it and points users to newer models.
How it came about
Mistral AI announced Mistral Large as its new flagship, alongside a faster, cheaper sibling called Mistral Small. The model went live on the company's own platform, la Plateforme, and on Microsoft Azure, which Mistral AI named as its first distribution partner. Beta customers on Azure were said by the company to have used it successfully. Mistral AI later marked the model as deprecated and replaced it with more capable releases.
Inside the model
The company describes Mistral Large as a text generation model tuned for reasoning, with strong instruction-following that developers can use to build their own moderation rules; Mistral AI used it for system-level moderation in le Chat. It supports function calling, so it can be wired to outside tools, APIs and databases, and a constrained output mode on la Plateforme. A JSON format mode makes the model emit valid JSON for structured pipelines. Parameter count, training data and hardware were not published.
What it can do
Mistral AI reported benchmark results for the pre-trained model on MMLU, HellaSwag, WinoGrande, ARC-Challenge, TriviaQA and TruthfulQA, and on coding and maths sets including HumanEval, MBPP, Math and GSM8K. The company claimed a second-place standing among models generally reachable through an API, behind GPT-4. In French, German, Spanish and Italian it was reported to beat LLaMA 2 70B clearly on HellaSwag, ARC-Challenge and MMLU. No limits were described.
How to use it
The model was served through la Plateforme, hosted on Mistral AI infrastructure in Europe, and through Azure AI Studio and Azure Machine Learning. It also powered the le Chat assistant demonstrator. Self-deployment with model weights was offered for sensitive cases on request. It is now deprecated and no longer maintained.
Safety and testing
Precise instruction-following let developers set their own moderation policies. Mistral AI says it used the model for system-level moderation inside le Chat. A constrained output mode on la Plateforme restricts what the model can produce. Nothing further on evaluations or incidents was published.
Reception and debate
Mistral AI positioned the model as the strongest one openly reachable through an API apart from GPT-4, based on its own benchmark figures. Beta users on Azure were described by the company as having had significant success. No independent reviews, disputes or bans were reported.
Why it matters
The release paired a flagship commercial model with open-weight endpoints, and Mistral AI framed its Azure deal with Microsoft as a step toward making frontier AI widely available. Function calling and JSON output were introduced on this model and the matching Mistral Small, with plans to extend formatting controls to the rest of the endpoint line-up.
Questions people ask about Mistral Large
What was Mistral Large built to do?
Mistral AI designed it for demanding reasoning, text understanding and rewriting, and code generation. It also handles multilingual work in English, French, Spanish, German and Italian, and can call external tools through function calling.
How can I use it now?
You cannot. Mistral AI has marked Mistral Large as deprecated, saying it is no longer maintained and has been superseded by newer, stronger models from the company.
How big is the model?
No parameter count was published. What is known is a 32,000-token context window, which sets how much text it can read and draw on at once.
Where did it run?
It was served from Mistral AI's la Plateforme, hosted on European infrastructure, and from Microsoft Azure via Azure AI Studio and Azure Machine Learning. Sensitive deployments could be self-hosted with weights on request.
Good reading and sources
There is no English Wikipedia article on Mistral Large yet, so the facts on this page come from the sources listed above: mistral.ai/news/mistral-large. Some are the subject's own pages, so its claims about itself are reported as its claims. EveryLLM wrote the text in its own words, with AI assistance, and compares it with the sources to make sure no passage is copied.
It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.