Snowflake opens public preview of Cortex Training for open-weight models

Snowflake announced the public preview of Cortex Training, a service for customizing and deploying open-weight language models without managing GPU clusters. Teams can use supervised fine-tuning, which teaches from examples, or reinforcement learning, which trains from feedback on outcomes, through existing frameworks or lower-level APIs. Snowflake manages GPU provisioning, distributed execution and checkpoints while keeping training data, model weights and optimizer state in its managed environment. The service supports full-parameter and LoRA training and deployment through OpenAI-compatible inference endpoints. Usage is billed by GPU time.