Weights & Biases

Weights & Biases builds tools for AI developers to train and manage models and develop AI applications. W&B Models supports model development and management, while W&B Weave helps teams trace, evaluate and monitor agents and applications. Its platform also offers serverless reinforcement learning for fine-tuning, hosted model inference, and a registry for datasets, models, prompts and code. Teams can choose SaaS, dedicated or customer-managed deployments.

Founded in 2017 by Lukas Biewald, Chris Van Pelt and Shawn Lewis, the company began with experiment tracking developed to help build models at OpenAI. Biewald now serves as general manager; Van Pelt and Lewis remain distinguished engineers. Its engineering approach makes experimental work inspectable: Weave instruments application functions to capture inputs, outputs and trace trees, and supports comparable evaluations across language-model use cases. Developers can follow application behavior from experimentation through production.

CoreWeave completed its acquisition of Weights & Biases in May 2025, with acquisition-date purchase consideration of $1.029 billion. The deal added model-training and AI application development tools to CoreWeave’s software offering. In June 2026, CoreWeave reported nearly one billion cumulative runs and trillions of metrics tracked in Weights & Biases, giving a measure of the experimental workloads handled by the platform.