| Mistral AI SAS | |
|---|---|
| Type | Private |
| Industry | Artificial intelligence |
| Founded | 28 April 2023 |
| Founders |
|
| Headquarters | , France |
| Key people |
|
| Products |
|
Number of employees | 150 (2025) |
Mistral AI SAS is a French artificial intelligence (AI) company based in Paris, specializing in large language models (LLMs) and generative AI technologies. Founded on 28 April 2023 by Arthur Mensch, Guillaume Lample, and Timothée Lacroix, Mistral AI focuses on open-weight and efficient AI models to provide alternatives to proprietary systems. Named after the mistral, a strong wind in southern France, the company aims to bring clarity and innovation to the AI field.
Mistral AI’s models, such as Mistral 7B, Mixtral 8x7B, and Mistral Large 2, are used by businesses, developers, and researchers for tasks like text generation, coding, and data analysis. The company’s chatbot, Le Chat, and its open-source approach have positioned it as a key player in the global AI landscape, competing with companies like OpenAI, Anthropic, and Meta AI.
Mistral AI was founded by three AI researchers who met at École Polytechnique: Arthur Mensch, former researcher at Google DeepMind; Guillaume Lample, and Timothée Lacroix, both previously with Meta AI. Their goal was to create efficient, open-weight AI models to democratize access to AI technology and challenge the dominance of proprietary systems.
In June 2023, Mistral AI raised €105 million from investors, including Lightspeed Venture Partners, Eric Schmidt, and Xavier Niel, valuing the company at €240 million. By December 2023, a second funding round of €385 million, led by Andreessen Horowitz, BNP Paribas, and Salesforce, increased its valuation to €2 billion. In June 2024, a €600 million round, led by General Catalyst, raised the valuation to €5.8 billion, making Mistral AI a leading AI company outside the San Francisco Bay Area.
On 26 February 2024, Mistral AI partnered with Microsoft to integrate its models into Microsoft Azure, expanding its global reach. By 2025, the company employed 150 people and continued to develop new models and services.
Mistral AI promotes open-source innovation and accessibility in AI development. Unlike proprietary models from companies like OpenAI, Mistral releases many of its models under open licenses, such as Apache 2.0, to encourage research and collaboration. The company aims to make AI efficient, cost-effective, and widely available, positioning itself as a European alternative to U.S.-based AI giants.
Le Chat (pronounced \lə ʃa\ in French) is Mistral AI’s conversational AI assistant, launched in February 2024 as a competitor to ChatGPT. On 19 November 2024, it was updated to include:
On 6 February 2025, Le Chat was released as a mobile app for iOS and Android, offering web search, image generation, and real-time updates. A Pro subscription tier, priced at $14.99 per month, provides access to advanced models, unlimited messaging, and enhanced features.
Mistral AI develops a range of LLMs, from lightweight models like Mistral 7B to high-capacity models like Mistral Large 2. Most models are released with open weights under licenses like Apache 2.0, while some, like Mistral Large, are proprietary. Below is a summary of key models:
| Name | Release Date | Status | Parameters (Billion) | License | Notes |
|---|---|---|---|---|---|
| Mistral Small 3.1 | March 2025 | Active | 24 | Apache 2.0 | Efficient model with image understanding, leading small models category. |
| Mistral Large 2 | November 2024 | Active | 123 | Mistral Research License | High-performance model, competitive with Llama 3.1 405B, fluent in multiple languages. |
| Pixtral Large | November 2024 | Active | 124 | Mistral Research License | Enhanced vision-language model with 1-billion-parameter visual encoder. |
| Mixtral 8x22B | April 2024 | Active | 141 | Apache 2.0 | Sparse mixture-of-experts model, released via BitTorrent, high performance. |
| Codestral 22B | May 2024 | Active | 22 | Mistral Non-Production License | Code-focused model, fluent in over 80 programming languages. |
| Mathstral 7B | July 2024 | Active | 7 | Apache 2.0 | STEM-focused model, strong in math and science tasks. |
| Mistral 7B | September 2023 | Active | 7.3 | Apache 2.0 | First model, efficient with grouped-query attention, outperforms larger models. |
| Mixtral 8x7B | December 2023 | Active | 46.7 | Apache 2.0 | Mixture-of-experts model, outperforms GPT-3.5 in benchmarks. |
Mistral AI’s models are designed for efficiency and high performance, often rivaling larger models from competitors.
A 2024 study by Patronus AI found Mixtral 8x7B generated copyrighted text in 22% of responses when prompted with protected book content, compared to 44% for GPT-4.
Mistral AI’s models are used across industries for various purposes:
Mistral AI’s partnership with Microsoft, announced on 26 February 2024, integrates its models into Microsoft Azure, making them accessible to Azure’s global user base. This collaboration also supports Le Chat’s deployment as a multilingual conversational assistant. Additional partnerships with Black Forest Labs enhance Le Chat’s image generation capabilities.
Mistral AI has strengthened Europe’s position in the global AI race, ranking fourth among AI companies worldwide and first outside the San Francisco Bay Area, ahead of Cohere, Hugging Face, and Perplexity AI. Its open-weight models have fostered innovation by enabling developers and researchers to build on its technology, contributing to advancements in AI accessibility.
However, challenges remain. The company faces competition from larger U.S.-based firms, and its open-source approach raises concerns about misuse, such as generating harmful content or copyrighted material. Additionally, maintaining high performance while scaling models to handle diverse languages and tasks requires significant resources.
Mistral AI plans to:
These efforts aim to solidify Mistral AI’s role as a leader in open and efficient AI development.