Mistral Forge: Custom AI to Challenge OpenAI and Anthropic

Mistral Forge: A Response to the Challenges of AI in Business

In the business world, many artificial intelligence projects fail not due to a lack of technology, but because the models used are not suited to the specific needs of companies. These models, often trained on publicly available data from the internet, do not take into account internal documents, workflows, and institutional knowledge accumulated over the years.

It is in this context that Mistral, a French startup specializing in artificial intelligence, identified a significant opportunity. On Tuesday, it announced the launch of Mistral Forge, an innovative platform that allows companies to create custom AI models trained on their own data. This announcement was made during the Nvidia GTC, an annual technology conference that this year places a particular emphasis on AI and agentic models for businesses.

A Business-Focused Strategy

Mistral distinguishes itself from competitors like OpenAI and Anthropic by primarily focusing on the enterprise market. According to Arthur Mensch, CEO of Mistral, this strategy is paying off, with the company on track to achieve over $1 billion in annual recurring revenue this year. This focus on the enterprise sector aims to provide companies with greater control over their data and AI systems.

Elisa Salamanca, product manager at Mistral, explains that Forge enables businesses and governments to customize AI models according to their specific needs. Unlike other approaches that merely adjust existing models or add proprietary data through techniques like retrieval-augmented generation (RAG), Forge offers the possibility to train models from scratch.

Unique Advantages for Businesses

This ability to train custom models could address some limitations of current methods, particularly regarding the management of non-English or highly domain-specific data. Forge also provides greater control over model behavior, allowing companies to reduce their reliance on third-party model providers. This could also enable businesses to train agentic systems using reinforcement learning, thus avoiding risks such as changes or depreciation of models.

Timothée Lacroix, co-founder and CTO of Mistral, emphasizes that Forge utilizes Mistral's extensive library of open-weight AI models, including models like Mistral Small 4. This allows companies to customize these models to highlight certain aspects while downplaying others. "The trade-offs we make when building smaller models are that they simply cannot be as good on all topics as their larger counterparts, so the ability to customize them allows us to choose what we emphasize and what we leave aside," Lacroix stated.

Enhanced Technical Support

Mistral supports its clients in selecting the models and infrastructure to use, while leaving these decisions in their hands. For companies requiring additional support, Mistral's team of engineers integrates directly with clients to optimize data usage and tailor models to their specific needs. This support model is inspired by large companies like IBM and Palantir.

Elisa Salamanca clarifies that "As a product, Forge already comes with all the tools and infrastructure needed to generate synthetic data pipelines. But understanding how to build the right evaluations and ensuring you have the right amount of data is something that companies generally lack the expertise for, and that's what the deployed engineers bring to the table."

Strategic Partnerships

Mistral Forge is already being used by partners such as Ericsson, the European Space Agency, the Italian consulting firm Reply, as well as Singapore's DSO and HTX. Early users also include ASML, the Dutch chip manufacturer that led Mistral's Series C funding round last September at a valuation of €11.7 billion (approximately $13.8 billion at the time).

These collaborations highlight the main use cases for Forge, particularly for governments that need to adapt models to their language and culture, financial players with high compliance requirements, manufacturers with customization needs, and tech companies that must adjust models to their codebase. According to Marjorie Janiewicz, Mistral's revenue director, these sectors particularly benefit from Forge's ability to tailor AI models to specific languages, cultures, and needs.