Mistral AI Unveils Forge: Enterprise Custom AI Model Training Platform

James Ding
Mar 17, 2026 21:40

Mistral AI introduces Forge, empowering enterprises to train cutting-edge AI models on internal data. ASML, Ericsson, ESA, among launch partners.

Mistral AI has launched Forge, a groundbreaking platform that allows enterprises to develop AI models trained exclusively on their internal data. This innovative approach eliminates the limitations of fine-tuning public models, revolutionizing corporate AI adoption.

The French AI company has secured partnerships with industry giants such as ASML, Ericsson, the European Space Agency, and Singapore’s DSO National Laboratories and Home Team Science and Technology Agency. These partners will leverage Forge to train models on proprietary datasets that drive their most critical operations.

Revolutionizing AI Deployment

Unlike traditional enterprise AI deployments that rely on fine-tuning public models with limited internal data, Forge supports the entire training lifecycle. It enables pre-training on extensive internal datasets, post-training refinement, and reinforcement learning to ensure alignment with company policies.

The platform accommodates dense and mixture-of-experts (MoE) architectures, offering superior performance while reducing latency and compute costs—a crucial benefit for enterprises managing AI infrastructure budgets.

Mistral emphasized that generic models trained on public web data fail to capture the internal knowledge, operational processes, and institutional decisions unique to each enterprise.

Agent-Centric Design

Forge is designed with autonomous AI agents as the primary users, streamlining model fine-tuning, hyperparameter optimization, training job scheduling, and synthetic data generation through intuitive English instructions. This agent-first approach empowers enterprise agents with the ability to navigate internal systems, make informed decisions, and execute complex workflows efficiently.

Targeted Applications

Mistral identified specific use cases for Forge, including financial institutions training on compliance frameworks, software teams developing models for proprietary codebases, manufacturers creating diagnostics models, and government agencies analyzing policies across various regulatory frameworks and languages.

Organizations can continuously improve models through reinforcement learning pipelines, ensuring adaptability to evolving regulations, system changes, and emerging data.

Strategic Advancements

The launch of Forge coincides with Mistral’s release of the Mistral Small 4 model, Leanstral (an open-source code agent for formal verification), and collaboration with Nvidia’s Nemotron Coalition to develop the first open frontier base model.

Forge addresses concerns around AI infrastructure investments, providing enterprises with control over models, training data, and intellectual property. Models remain within the enterprise environment, governed by internal policies rather than third-party terms of service.

Details on pricing and availability are forthcoming, and interested organizations can register for early access on Mistral’s website.

Image source: Shutterstock