Loading...
Loading...
Mistral AI announced Forge, an enterprise model training platform that lets organizations build, customize, and continuously improve proprietary AI models using their own data, positioning the French lab against hyperscale cloud providers. Forge supports full lifecycle training—pre-training on internal data, supervised fine-tuning, DPO/ODPO, and reinforcement learning pipelines—to align models with company policies and operational goals. Mistral says Forge packages the company’s internal trainin
Mistral AI Releases Forge | Hacker News Hacker News new | past | comments | ask | show | jobs | submit login Mistral AI Releases Forge ( mistral.ai ) 33 points by pember 2 hours ago | hide | past | favorite | discuss help Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact Search:
Mistral AI today launched Forge, a platform for enterprises to train frontier-grade AI models on proprietary data so they reflect internal knowledge, workflows, and policies. Forge supports pre-training, post-training, and reinforcement learning on internal docs, codebases, structured data and operational records, enabling agents that use correct terminology, follow procedures, and make compliant decisions. Mistral says customers include ASML, Ericsson, ESA, Singapore DSO and HTX, and Reply. The offering emphasizes control and strategic autonomy: models remain governed by an organization’s policies and can be run within enterprise infrastructure to meet regulatory and IP requirements. Forge aims to improve tool use, multi-step orchestration and domain-aligned decision-making for production agents.
Mistral AI today launched Forge, a platform for enterprises to train “frontier-grade” models on proprietary data so AI reflects internal knowledge, policies, codebases and workflows. The company says customers including ASML, Ericsson, the European Space Agency and Singapore’s DSO and HTX are already using Forge to pre-train, fine-tune and apply reinforcement learning on internal datasets. Forge aims to give organizations control and strategic autonomy by keeping models and IP governed by internal policies and infrastructure, and to produce more reliable enterprise agents that select tools correctly, follow procedures and comply with regulations. The product targets regulated and complex engineering, operations and decision-making environments.
Mistral AI announced Forge, an enterprise model training platform that lets organizations build, customize, and continuously improve proprietary AI models using their own data, positioning the French lab against hyperscale cloud providers. Forge supports full lifecycle training—pre-training on internal data, supervised fine-tuning, DPO/ODPO, and reinforcement learning pipelines—to align models with company policies and operational goals. Mistral says Forge packages the company’s internal training recipes, data mixing, generation pipelines, and distributed compute optimizations that go beyond simple fine-tuning APIs offered by OpenAI, Anthropic, Google and others. Early customers cite niche needs where off-the-shelf models fall short, signaling demand for deeper customization despite higher cost and complexity.