Mistral Forge

Video Review

About

Mistral Forge is an enterprise AI training platform that lets organizations build custom large language models trained exclusively on their own proprietary data — without routing sensitive information through shared cloud infrastructure. Launched at Nvidia GTC in March 2026, Forge packages the exact training methodology Mistral's own scientists use internally to build production models. The platform covers the full model training lifecycle: pre-training on large internal datasets, post-training via supervised fine-tuning (SFT), direct preference optimization (DPO), and reinforcement learning pipelines aligned to internal evaluation criteria. Data mixing strategies, distributed computing optimizations, and battle-tested training recipes are all included. Forge ships with structured function calling support essential for agentic applications — models can be taught to query databases, call external APIs, and take actions based on business rules, not generic training data. Mistral's team of forward-deployed engineers embeds with enterprise customers to surface the right datasets and adapt pipelines to specific use cases. Early adopters include Ericsson, the European Space Agency, Italian consulting firm Reply, and Singapore's Defense Science and Technology Agency (DSO) and Home Team Science and Technology Agency (HTX). For organizations where data residency, IP protection, or regulatory compliance make generic cloud models unacceptable, Forge offers a path to building AI that actually belongs to you.

Key Features

• Pre-training on proprietary datasets at scale
• Post-training: SFT, DPO, ODPO, and RL alignment pipelines
• Distributed computing with data mixing and training recipe optimization
• Structured function calling for agentic applications
• Forward-deployed Mistral engineers embedded with your team
• Full data residency — your data never leaves your infrastructure
• Self-evolving models with ongoing RL fine-tuning

Use Cases

• 1Building proprietary LLMs trained on internal knowledge bases
• 2Regulatory compliance (HIPAA, GDPR, sector-specific data laws)
• 3Defense and government AI systems requiring data sovereignty
• 4Enterprise AI agents with domain-specific function calling
• 5Custom model training without third-party data exposure