Weights & Biases
MLOps platform for tracking experiments, visualizing training runs, and collaborating on AI projects
Overview
Weights & Biases (W&B) is an MLOps platform for tracking experiments, visualizing training metrics, and collaborating on machine learning and LLM projects. The Free plan allows up to 5 users with 5 GB of monthly storage and 1 GB of ingestion for the Weave evaluation tool. The Pro plan starts at $60/month, reserved for teams with fewer than 50 employees, with 100 GB of included storage ($0.03/GB additional), 10 seats, and team-level access controls. The Enterprise plan is by quote, with single tenant, HIPAA compliance, SSO, and audit logs. A free self-hosted Personal version exists for a single user, and an advanced self-hosted Enterprise offering is also available by quote.
Weights & Biases is not translated into French: the interface and documentation are in English only. The vendor does not document guaranteed European hosting for standard cloud plans; only the Enterprise offering with single-tenant option allows you to choose your region. An API and Python SDKs are provided, with official integrations to PyTorch, TensorFlow, Hugging Face, Keras, scikit-learn, and OpenAI. Qualified academic researchers benefit from free Pro-level access with 200 GB of storage. The main limitation is the cost of Weave data ingestion ($0.10/MB beyond the Pro plan), which can escalate quickly for teams tracing many LLM calls in production.
Our verdict
Best for research and data science teams wanting to visually compare dozens of experiments and collaborate without reinventing a dashboard. Not for you if you are tracing high volumes of LLM calls in production without monitoring costs: Weave ingestion beyond the Pro plan is billed per megabyte and can quickly exceed budget.