Wedoany.com Report on Mar 18th, French artificial intelligence company Mistral AI launched the Forge platform, a training platform that allows enterprises to build, customize, and continuously improve AI models using their own proprietary data. This move signifies the company's direct competition with major cloud service providers in the enterprise technology sector.

In an interview, Mistral AI's Head of Product, Elisa Salamanca, stated: "Forge is Mistral's model training platform. We have been building this platform behind the scenes with AI scientists. The real value that Forge brings is that it enables businesses and governments to customize AI models according to their specific needs."

The platform supports the complete model training lifecycle, including pre-training, post-training, and reinforcement learning processes. Salamanca noted that Forge packages the training methodologies Mistral uses internally to build its flagship models, including data blending strategies and proven training recipes.

Explaining why enterprises need to train their own models from scratch, Salamanca said: "Many existing models can take you quite far. But when you consider what makes you competitive against your rivals—everyone can adopt and use existing models. When you want to go a step further, you actually need to create your own model. You need to leverage your proprietary information."

She provided examples, stating that Mistral had collaborated with a public institution to create a custom model for filling in missing parts of ancient manuscripts and helped Ericsson customize a code translation model. In the financial sector, Mistral assisted a hedge fund in developing a unique model based on a proprietary quantitative language.

Forge's business model includes platform licensing fees, data pipeline service fees, and embedded AI researcher services. Salamanca emphasized: "Currently, no competitor sells this kind of embedded scientist as part of their training platform offering."

Data privacy is a key differentiator for Forge. Salamanca stated: "It's on their cluster, using their data—we see nothing, so it's entirely under their control. I think this sets us apart from the competition."

The platform also supports an "agent-first" design, allowing autonomous agents to initiate training experiments. Salamanca believes that even within agent architectures, model customization is essential: "You need to bring some agent behavior to the model. This could involve reasoning patterns, specific types of documents, ensuring you have the correct reasoning traces."

In the same week, Mistral also launched the Leanstral code agent and the Mistral Small 4 model, and participated as a co-developer in the Nvidia Nemotron Alliance for developing open frontier models. Mistral AI co-founder and CEO Arthur Mensch said: "Open frontier models are how AI becomes a true platform. Together with Nvidia, we will play a leading role in training at scale and advancing frontier models."

Forge faces competition from platforms like Amazon Bedrock and Microsoft Azure AI Foundry, but Salamanca pointed out that these products are primarily limited to the cloud and lack deep control. She warned of the risks of relying on closed-source models: "When you rely on closed-source models, you are also super dependent on model updates, which have side effects."

Mistral positions Forge as an open-source platform and plans to support other open-source architectures in the future. Salamanca confirmed: "We are deeply rooted in open source. This has been part of our DNA from the beginning, and we have been building Forge as an open platform—it's only a matter of time before it opens to other open-source models."

The platform's launch comes as Mistral faces talent competition, but the company is building lasting competitive assets through institutional capabilities and platformized expertise. Forge represents a bet on the future of enterprise AI: systems trained on proprietary knowledge, governed by internal policies, and operating under the organization's direct control will be the most valuable.