Weights and Biases overview: experiment tracking, model registry, and LLM evaluation with Weave. Pricing plans, key features, pros, cons, and FAQs.
Weights and Biases is an AI developer platform used by machine learning and generative AI teams to track experiments, manage model versions, and evaluate AI applications. It was founded in 2017 by Lukas Biewald, Chris Van Pelt, and Shawn Lewis and is headquartered in San Francisco.
What began as an experiment tracking tool for model training has grown into a broader MLOps platform, now spanning traditional model development as well as large language model application evaluation through its Weave product line.
The platform is used by both individual researchers and enterprise ML teams, with deep integrations into the tools and frameworks that make up a typical machine learning workflow.
Experiment tracking lets teams log metrics, hyperparameters, and system information during training, then compare runs visually to understand what changes improve model performance.
The Model Registry adds versioning, lineage tracking, and governance for models and datasets, while Weave brings tracing, evaluation, and scoring capabilities purpose-built for LLM and GenAI applications.
Sweeps automate hyperparameter optimization, and native integrations with PyTorch, TensorFlow, Hugging Face, OpenAI, and LangChain let teams adopt the platform without changing their existing stack.
A free tier is available for individual and academic use, including unlimited projects and teams and 5GB of monthly storage, though it prohibits corporate use.
The Pro plan starts at $60 per month and targets early-stage teams under 50 employees, adding more model seats, unlimited team collaboration, priority support, and larger storage and data ingestion allowances, with overage storage billed at $0.03 per gigabyte.
Enterprise plans are custom-priced and billed annually, adding single-tenant deployment, HIPAA compliance, SSO, custom roles, and audit logs, and a self-hosted deployment option is available for organizations with stricter data residency needs.
Yes, there is a free tier for individuals and academic use with 5GB of monthly storage, but corporate use requires a paid plan.
Lukas Biewald, Chris Van Pelt, and Shawn Lewis founded the company in 2017.
Weave is a toolkit within the platform for tracing, evaluating, and monitoring LLM and GenAI applications.
Yes, self-hosted and single-tenant enterprise deployment options are available.
PyTorch, TensorFlow, Keras, Hugging Face, scikit-learn, OpenAI, LangChain, and other common ML tools.
Yes, a free Pro-tier academic license is available for qualifying students and researchers.
Weights and Biases became part of CoreWeave.
Yes, through built-in Sweeps for automated hyperparameter optimization.