Weights & Biases
Weights & Biases (W&B) tracks and manages AI model training. See 2026 pricing tiers, free plan limits, Weave LLM tools, and the CoreWeave ownership change.
About Weights & Biases
What is Weights & Biases (2026 update)
Weights & Biases, usually called W&B, is an AI developer platform for training, fine-tuning, managing, and tracking machine learning models and AI applications. Teams use it to log experiments, compare model runs side by side, store datasets and model versions in a registry, and trace what a live AI application actually did when it produced a given output. In May 2025, cloud compute provider CoreWeave completed its roughly 1.7 billion dollar acquisition of W&B, and as of 2026 the two products are being sold and marketed as one combined AI cloud and developer platform.
What is new in 2026
The CoreWeave ownership change is the headline update, since W&B is now positioned alongside CoreWeave’s GPU cloud rather than as a fully independent company. On the product side, W&B Weave has expanded into LLM tracing, evaluation, and guardrails, letting teams track prompts and outputs, flag toxicity, bias, and hallucinations, and compare model performance across different providers in one view. The Registry has gotten a faster type-ahead search, redesigned Registry and Collection cards, and new icons for visually distinguishing assets, with AI-assisted ranking described as coming soon. W&B has also shipped security updates to API key management for both personal and service accounts, and its self-hosted W&B Server line continues to release regular version updates, with 0.82.0 adding editable collection overviews and version-level metric charts.
Key features
- Experiment tracking that logs metrics, hyperparameters, and system stats for every training run
- Model and dataset registry with lineage tracking so teams can see what data and code produced a given model
- W&B Weave for tracing, evaluating, and guardrailing LLM-based applications in production
- Team collaboration tools including shared workspaces, reports, and CI/CD automations for ML pipelines
- Mobile app for monitoring runs, with starred metrics, cleaner run cards, and an improved log viewer
- Self-hosted W&B Server option for teams that need to run the platform on their own infrastructure
Pricing in 2026
W&B prices around model seats, storage, and Weave data ingestion rather than a flat per-user fee.
| Plan | Price | What you get |
|---|---|---|
| Free | Free | Up to 5 model seats, unlimited Weave seats, 5 GB/month storage, 1 GB/month Weave data ingestion, community support, personal use only |
| Pro | $60/month billed monthly or annually | Everything in Free, unlimited teams, service accounts, priority support, CI/CD automations, up to 10 model seats, 100 GB/month storage, 1.5 GB/month Weave ingestion, $5/month inference credit |
| Enterprise | Custom | Single-tenant deployment, HIPAA-compliant option, SSO, automated provisioning, custom roles and audit logs, customizable seats and storage |
| Academic | Free | Pro-level features for qualifying academic use, with 200 GB storage and up to 25 GB/month Weave ingestion |
Who should use it
W&B fits ML engineers and research teams who need a serious system of record for training runs and model versions instead of scattered spreadsheets and log files. The free plan works for individual researchers and students, the Pro plan suits early-stage startups under 50 employees, and larger organizations needing compliance features or private deployment should expect to move to Enterprise, since W&B explicitly requires teams that outgrow Pro’s guidelines to make that switch.
Limitations and alternatives
Pro plan pricing is capped by team size and storage, so fast-growing teams can hit those Enterprise-transition limits sooner than expected, and Enterprise pricing itself is not public, which makes budgeting harder to plan in advance. Being newly owned by a compute provider also means some users are watching to see how independent W&B’s roadmap stays over time. If W&B is not the right fit, commonly cited alternatives include MLflow for a self-hosted, open source experiment tracker, Comet ML for a similar hosted experiment tracking and model registry service, and Neptune.ai for teams that want lighter-weight experiment logging.
Frequently asked questions
Is Weights & Biases free to use?
Yes, W&B has a free plan with up to 5 model seats, unlimited Weave seats, and 5 GB/month of storage, but it is restricted to personal, non-corporate use.
How much does W&B Pro cost?
As of 2026, the Pro plan starts at $60/month, billed monthly or annually, and is designed for early-stage teams under 50 employees.
Is Weights & Biases still independent in 2026?
No. CoreWeave completed its acquisition of Weights & Biases in May 2025 for about 1.7 billion dollars, and the two are now marketed together as a combined AI cloud and developer platform.
What is W&B Weave used for?
Weave is W&B's toolset for LLM applications, covering tracing, evaluation, and guardrails, including detecting toxicity, bias, and hallucinations and comparing model performance across providers.
Information last verified: September 2026.
Related Tools
