Overview
- Announced at Nvidia's GTC, Forge is now in use with ASML, Ericsson, the European Space Agency, Reply, and Singapore's DSO and HTX.
- Companies can train models from scratch rather than only fine-tune or rely on RAG, targeting deeper grasp of internal terminology and workflows.
- The system supports dense and mixture-of-experts architectures, multimodal inputs, reinforcement learning pipelines, and continuous evaluation.
- Mistral provides forward-deployed engineers plus tooling for synthetic data generation, data surfacing, and enterprise-grade evals.
- CEO Arthur Mensch says the enterprise push is driving momentum, with guidance that annual recurring revenue will exceed $1 billion this year.