Mistral AI launches Forge, enterprise platform for training custom models on proprietary data

Mistral AI announced Forge on March 17, 2026, a platform enabling enterprises to train frontier-grade AI models on proprietary knowledge including codebases, compliance policies, engineering standards, and operational processes.

The company has already partnered with ASML, DSO National Laboratories Singapore, Ericsson, European Space Agency, Home Team Science and Technology Agency (HTX) Singapore, and Reply to train models on proprietary data.

Training capabilities

Forge supports three training approaches:

• Pre-training for building domain-aware models from large internal datasets
• Post-training for refining model behavior for specific tasks and environments
• Reinforcement learning for aligning models with internal policies and operational objectives

The platform supports both dense and mixture-of-experts (MoE) architectures. According to Mistral AI, MoE enables large models to run with lower latency and compute cost compared to dense models of similar scale. The system also supports multimodal inputs including text and images.

Agent-first design

Forge is designed for autonomous agents to use directly. Mistral AI claims its agent Mistral Vibe can use Forge to fine-tune models, optimize hyperparameters, schedule jobs, and generate synthetic data. The platform monitors metrics during training to prevent regression on specified benchmarks.

The system includes infrastructure management and data pipeline templates based on Mistral AI's training methods, allowing users to customize models through natural language instructions.

Enterprise applications

Mistral AI outlines several use cases:

• Government agencies training models on policy frameworks and regulatory texts in multiple languages and dialects
• Financial institutions building models on compliance frameworks and risk procedures
• Software teams training on proprietary codebases and development standards for implementation, debugging, and code review
• Manufacturers using engineering specifications and maintenance records for diagnostics and design analysis

Data control and governance

Forge allows organizations to train models using proprietary datasets governed by internal policies. Models can be operated within customer infrastructure environments. According to Mistral AI, this provides strategic autonomy for enterprises in regulated industries that must ensure models reflect compliance requirements and operational constraints.

The platform includes continuous improvement capabilities through reinforcement learning pipelines and evaluation frameworks for testing against internal benchmarks before production deployment.

What this means

Forge represents Mistral AI's move into the enterprise custom model training market, competing with offerings from OpenAI, Anthropic, and cloud providers. The agent-first design suggests Mistral AI is positioning for a future where autonomous systems handle model training and optimization. The early partner list includes organizations in semiconductors, defense, telecommunications, and aerospace—sectors with significant proprietary data and strict compliance requirements. Pricing and technical specifications for context windows, parameter counts, and training scale were not disclosed.