Mistral said on 17 March 2026 that it had introduced Forge, a system for enterprises to build frontier-grade AI models grounded in their proprietary knowledge.
Forge covers pre-training on large internal datasets, post-training for specific tasks, and reinforcement learning to align models and agents with internal policies, evaluation criteria and operational objectives. It supports dense and mixture-of-experts architectures and multimodal inputs where needed.
Mistral said it has already partnered with organisations including ASML, DSO National Laboratories Singapore, Ericsson, the European Space Agency, Home Team Science and Technology Agency (HTX) Singapore, and Reply to train models on proprietary data.
The platform is agent-first: an autonomous agent such as Mistral Vibe can fine-tune models, find hyperparameters, schedule jobs and generate synthetic data, while Forge monitors metrics against chosen benchmarks. Infrastructure and training recipes are included so customisation can be driven in plain English.
Example uses span government policy models, financial compliance, proprietary codebase assistants and manufacturing diagnostics. Mistral framed custom models as strategic assets that keep institutional knowledge under enterprise control.