Weights & Biases
AI developer platform for experiment tracking, sweeps, model registry and artifacts, plus Weave for LLM tracing and evals; part of CoreWeave
Overview
Weights & Biases (W&B) calls itself "the AI developer platform to build AI agents, applications, and models with confidence." Its Models product is the experiment tracker many ML teams standardize on: Experiments for logging runs, Sweeps for hyperparameter search, Tables, Artifacts for dataset and model lineage, a Registry and Reports for sharing results. Weave adds LLM and agent tracing, evaluations, a playground and monitors, and the platform also offers serverless inference, Launch and automations. Lukas Biewald, Chris Van Pelt and Shawn Lewis founded the company in San Francisco in 2017. CoreWeave announced its acquisition on March 4, 2025 and closed it on May 5, 2025; the site now says W&B "is now part of CoreWeave Forge," and pricing lives on CoreWeave's site. The open-source wandb SDK is MIT licensed (latest v0.31.0, October 2026). Also on this site: W&B Weave and CoreWeave.
The Verdict
Who Should Use Weights & Biases?
Best For
- Research and training teams that want polished run dashboards and reports without hosting anything
- Teams that track classic training runs and LLM apps in one place (Models plus Weave)
- Academic labs (free academic plan)
Not Ideal For
- Cost-sensitive teams at large scale (some move to self-hosted MLflow)
- Teams that need self-hosting without an Enterprise contract
- Companies with 50 or more employees wanting the Pro plan (Pro is limited to smaller teams)
What's Great
- Strong visualizations and team collaboration, rated highly on G2 for ease of use and monitoring
- Broad toolset: sweeps, artifacts, registry, reports and Weave for LLM apps
- Generous free tier and a free academic plan
- Used by AI labs and large companies, including OpenAI, NVIDIA, Meta, Microsoft and Toyota
Watch Out For
- Costs climb as experiments, seats and storage scale
- Self-hosting is an Enterprise feature
- UI can be slow with large projects, per some G2 reviewers
- Roadmap now tied to CoreWeave Forge after the 2025 acquisition
Pricing
View all features & details
Models
- Experiments (run logging and comparison)
- Sweeps (hyperparameter search)
- Tables, Artifacts and lineage
- Registry and Reports
Weave and Platform
- Weave traces, evaluations, playground, monitors
- Serverless inference
- Launch and automations
- MIT-licensed Python SDK
Real-World Usage
Named Customers
- OpenAI, NVIDIA and Cohere teams (about page)
- Logos: AstraZeneca, BMW, Canva, Meta, Microsoft, Perplexity, Samsung, Toyota, Wayve
- "Thousands of businesses and millions of practitioners"
Ownership
- CoreWeave acquisition announced March 4, 2025, closed May 5, 2025
- Now sold as part of CoreWeave Forge
How It Compares
| Feature | Weights & Biases | MLflow | Comet | ClearML |
|---|---|---|---|---|
| Model | Hosted SaaS, self-host on Enterprise | Open source, free | Hosted SaaS plus open-source Opik | Open source plus SaaS |
| Free tier | Up to 5 model seats | Unlimited (self-hosted) | 1 user | Up to 3 users |
| Paid entry | From $60 per month | Managed via clouds | $19 per user per month | $15 per user per month |
| LLM tracing | Weave | MLflow tracing | Opik | GenAI App Engine |
Sources: wandb.ai · W&B pricing · GitHub · Comet pricing · ClearML pricing
Is this your tool? Add a badge