====== Mistral AI ======

**Mistral AI** is a French artificial intelligence company founded in April 2023 in Paris by **Arthur Mensch** (CEO, ex-DeepMind), **Guillaume Lample** (Chief Scientist, ex-Meta), and **Timothee Lacroix** (CTO, ex-Meta). The company develops large language models and enterprise AI solutions, emphasizing open-source efficiency and cost-effectiveness. By early 2026, Mistral reached a valuation of approximately $13.8 billion and projects over $1 billion in annual revenue.((source [[https://mistral.ai/about|Mistral AI About]]))

===== Models =====

==== Open-Weight Models ====

  * **Mistral 7B**: Breakthrough open-source model that matched or outperformed Meta's larger Llama 2, accumulating over 3.4 million HuggingFace downloads. Optimized for consumer hardware
  * **Mixtral 8x7B**: Open-weight Mixture of Experts model delivering strong benchmark performance with efficient resource usage
  * **Mixtral 8x22B**: Larger MoE variant for more demanding tasks
  * **Les Ministraux** (October 2024): Edge-computing models designed for phones and robotics((source [[https://brief.bismarckanalysis.com/p/ai-2026-mistral-will-rise-as-compute|Mistral AI Analysis 2026]]))

==== Proprietary Models ====

  * **Mistral Large**: Flagship proprietary model released February 2024, positioned as a GPT-4 competitor
  * **Mistral Medium 3.1** (August 2025): Multimodal model achieving approximately 90% of Claude Sonnet 3.7 performance at lower inference cost
  * **Codestral**: Dedicated model for code generation and programming tasks

===== La Plateforme =====

**La Plateforme** is Mistral's developer API platform (branded with the French-language convention alongside the **Le Chat** consumer chatbot). It provides access to the full model lineup for developers and enterprises, with emphasis on cost-efficient inference. Enterprise offerings include custom solutions and dedicated deployment options.((source [[https://mistral.ai/about|Mistral AI]]))

===== Mistral Forge =====

Announced at NVIDIA GTC in March 2026, **Mistral Forge** enables enterprises to train models from scratch on proprietary data rather than relying solely on fine-tuning or RAG approaches. This positions Mistral for deeper enterprise integration with agentic systems and full data control.((source [[https://techcrunch.com/2026/03/17/mistral-forge-nvidia-gtc-build-your-own-ai-enterprise/|Mistral Forge at NVIDIA GTC]]))

===== EU AI Focus =====

Mistral has strong backing from French government leadership including President Macron, and positions itself as Europe's leading AI alternative to American tech giants. The company leverages EU support to build sovereign AI infrastructure, excelling in cost-efficient, multilingual, and domain-specific models.

===== Funding =====

^ Round ^ Amount ^ Valuation ^ Notable Investors ^
| Early rounds | Undisclosed | - | Various |
| Series C (Sep 2025) | 1.7 billion EUR | 11.7 billion EUR | ASML (11% stake), NVIDIA |

ASML became the largest shareholder at 11%. Total funding exceeds $3 billion. The company had approximately 350 employees by September 2025, many recruited from Meta's Llama team.((source [[https://mistral.ai/news/mistral-ai-raises-1-7-b-to-accelerate-technological-progress-with-ai|Mistral AI Series C]]))

===== See Also =====

  * [[deepseek|DeepSeek]]
  * [[cohere|Cohere]]
  * [[anthropic|Anthropic]]

===== References =====