Mistral AI

Mistral AI was established by three French AI researchers — Arthur Mensch (former Google DeepMind), Guillaume Lample and Timothée Lacroix (former Meta) — who originally met at École Polytechnique.

Mistral AI raised €105 million ($117 million) in seed funding from Lightspeed Venture Partners, Eric Schmidt, Xavier Niel, and JCDecaux, with a valuation of approximately €240 million.

Mistral AI released Mistral 7B, a 7-billion-parameter open-weight model that outperformed larger models on benchmarks, establishing the company's technical credibility and open-source approach.

Mistral AI released Mistral 7B via BitTorrent without a formal announcement, demonstrating performance on par with Llama 2 13B across all benchmarks while being significantly smaller. The surprise release established Mistral's reputation for engineering efficiency and sparked widespread adoption in the open-source community.

Mistral raised €385 million ($428 million) in Series A from Andreessen Horowitz, BNP Paribas, and Salesforce. In February 2024, Microsoft invested $16 million, making Mistral models available on Azure.

Mistral AI released Mixtral 8x7B, a sparse mixture-of-experts model that matched or exceeded GPT-3.5 and Llama 2 70B performance while using only 12.9B active parameters per token. The release validated MoE architectures for open-weight models and became one of the most influential open-source LLMs of 2024.

Mistral AI launched Le Chat, its consumer-facing chatbot platform, with a web interface and mobile apps featuring Mistral's latest models. The launch marked Mistral's transition from a model-only provider to a direct consumer AI product company competing with ChatGPT and Claude in the European market.

Mistral secured €600 million ($645 million) in funding, increasing its valuation to €5.8 billion ($6.2 billion), ranking fourth globally in the AI industry by valuation and first outside the San Francisco Bay Area.

Mistral AI released Mistral Large 2, its flagship model with strong multilingual performance across French, German, Spanish, Italian, and English alongside coding and reasoning improvements. The release solidified Mistral's position as Europe's leading AI lab and demonstrated competitive frontier-model capabilities outside the US-China axis.

Mistral AI released Le Chat, its consumer chatbot, on iOS and Android. The app included image generation using Black Forest Labs' Flux Pro model and competed with ChatGPT and Claude.

Mistral secured €2 billion investment valuing it at €12 billion ($14 billion), with $1.5 billion from ASML which owns 11% of Mistral. The round highlighted growing European investment in AI sovereignty.

Mistral AI released Mistral 3, its next-generation model family including Mistral Large 3, a 675 billion parameter sparse mixture-of-experts model with 41B active parameters, and three smaller dense models (Ministral 3 at 14B, 8B, and 3B). All models were released under the Apache 2.0 license with open weights. Mistral Large 3 was trained from scratch on 3,000 NVIDIA H200 GPUs and debuted as the second-best open-weight non-reasoning model on the LMArena leaderboard. The release marked Mistral's return to MoE architecture since the Mixtral series and represented a substantial step forward in pretraining capability.

Mistral AI launched Voxtral Transcribe 2, a state-of-the-art speech-to-text model with sub-200ms latency, native diarization support, and open weights under Apache 2.0. The model was widely recognized as the first to dethrone OpenAI's Whisper after roughly three years as the open-source speech recognition leader, marking Mistral's entry into voice AI infrastructure.

Mistral AI acquired Paris-based AI startup Koyeb in February and raised $830 million in March to build new data centers near Paris and in Sweden. This positioned Mistral as Europe's sovereign AI infrastructure champion.

Mistral AI released Small 4, a 119B-parameter Mixture-of-Experts model with 128 experts and only 6B active parameters per token, released under Apache 2.0. It unified the previously separate Pixtral (vision), Devstral (coding), and Magistral (reasoning) model lines into a single architecture with configurable reasoning effort, deployable on a single H100 when compressed.

Mistral AI launched Medium 3.5, a 128B dense flagship model with 256K context window and configurable reasoning, released with open weights on Hugging Face. Alongside it, Mistral shipped Vibe, an IDE-style coding agent that pairs with Medium 3.5 for autonomous code generation. The release also included Le Chat Enterprise, marking Mistral's push into enterprise AI productivity.

Mistral AI announced the acquisition of Austrian company Emmi AI, which develops AI simulation models for industrial engineering. This expanded Mistral's capabilities beyond language models into physical and industrial AI applications.

Mistral released Leanstral 1.5, a free Apache-2.0 licensed model with 119B total (6B active) parameters for formal verification in Lean 4. The model saturated miniF2F benchmark, solved 587/672 PutnamBench problems, and achieved state-of-the-art results on FATE-H (87%) and FATE-X (34%). It uncovered 5 previously unknown bugs across 57 open-source repositories tested, marking a significant advance in practical AI-assisted formal verification.

Mistral AI introduced Robostral Navigate, an 8B-parameter vision-language model for embodied robot navigation that uses only a single RGB camera — no LiDAR, depth sensors, or multi-camera rigs needed. Achieves 76.6% success on R2R-CE benchmark, marking Mistral's expansion into physical AI and robotics.

Mistral announced a EUR 3 billion Series D round at a post-money valuation above EUR 21 billion, the largest equity fundraising ever completed by a European technology company. Samsung Electronics led the round, with Scaleup Europe Fund (EQT) and PSG Equity as co-leads; new investors included Advent, BlackRock funds, and the Grand Duchy of Luxembourg, alongside existing backers NVIDIA, ASML, a16z and General Catalyst. Three years after its seed round, Mistral pledged the capital to scale frontier research and sovereign compute capacity, cementing its strategy as Europe's full-stack open-weight alternative to US and Chinese AI labs.