Mistral AI releases Mistral Large 4 to challenge rivals

What is Mistral Large 4?

Mistral Large 4 is a multimodal model, meaning it can process multiple types of data, such as text and images. With 1 trillion parameters, it is a highly capable system built to compete directly with proprietary, closed-source models.

Unlike closed models that are controlled entirely by their creators, Mistral plans to make the "weights"—the core blueprint and settings of the model—freely available to the public. This allows businesses to run and customize the model on their own systems.

Currently, the model is accessible in a preview phase through a public guardrail endpoint. Mistral plans to release the open-weight version in three weeks, once safety testing is complete. The company is working with trusted partners and governments to ensure the open-source weights are used for defense rather than malicious attacks.

How was the model trained?

Mistral trained Mistral Large 4 entirely on its own computing infrastructure. The company used 4,000 Nvidia graphics processing units (GPUs), which are specialized computer chips used to train AI. According to Mistral, this is two to three times fewer chips than its Chinese competitors used, and significantly less than what closed-source American rivals require.

Unlike some Chinese labs that have been accused of using "distillation"—a process where a smaller model is trained on the outputs of a larger one—Mistral claims it trained Mistral Large 4 completely from scratch.

What are the key business use cases?

While Mistral Large 4 is designed to compete as a general-purpose model, it has been optimized for specific industries and technical tasks. These specialized areas include:

Why it matters for businesses

For businesses, choosing open-weight models like Mistral Large 4 offers significant advantages. Running open-weight models is often much cheaper because companies only pay for the computer power they consume, rather than paying ongoing subscription fees to external providers.

Additionally, using an open-weight model gives businesses complete control over their technology. With closed-source models, companies face the risk that a provider could suddenly change, restrict, or revoke access to the AI. This risk became clear recently when the US government placed temporary restrictions on distributing certain models from OpenAI and Anthropic due to security concerns.

By owning and hosting the model themselves, businesses can ensure their AI-powered defenses and operations remain stable and secure, without relying on foreign tech giants.