French artificial intelligence lab Mistral AI SAS today announced that it has raised €3 billion, or about $3.49 billion, in funding.

Samsung Electronics Co. Ltd. led the investment. It was joined by more than two dozen other backers including Salesforce Ventures, Nvidia Corp. and ASML Holdings NV, which led Mistral’s last round. The startup is now valued at more than €20 billion.

It’s best known for its lineup of open-source large language models. Its flagship LLM, Mistral Medium 3.5, outperforms algorithms several times its size on popular coding benchmarks. The company also develops more specialized models. In August, it debuted a so-called guard model optimized for tasks such as protecting LLMs from malicious prompts.

Mistral makes money by selling managed cloud-based versions of its algorithms. Additionally, it offers development tools that enterprises can use to build their own custom LLMs.

Mistral Forge is a cloud service that enables developers to generate synthetic training datasets. Those are file collections made up of AI-generated records that would be difficult to create manually. After generating a dataset, Mistral Studio can use it to perform pretraining, the initial step of the LLM training workflow. The tool also manages the subsequent reinforcement learning phase, which hones the model’s reasoning skills.

A second cloud service called Mistral Studio enables companies to turn their custom LLMs into AI agents. Developers can connect an agent to data repositories, equip it with system prompts and add guardrails. An error mitigation mechanism enables agents to automatically retry actions that they fail to complete on the first try.

Mistral Studio includes a so-called AI registry. It’s a repository that developers can use to store their AI agents along with datasets, prompts and related assets. A version control module can roll back agent components to an earlier iteration if a bug finds its way into the code.

The company sells its agent development tools alongside a pre-packaged agent called Vibe. It can generate code and automate less technical work such as creating spreadsheets. One of Vibe’s main differentiators is that customers can run it on-premises, which eases regulatory compliance in some verticals.

Mistral disclosed today that it has more than 125 enterprise clients. The group includes major brands such as ASML, Airbus SE and HSBC Holdings plc. In April, Mistral forecasted that its monthly revenue would reach $80 million by year’s end. 

The company will use its new capital to purchase computing infrastructure for AI training projects. Mistral also plans to grow its international presence. 

Image: Mistral AI