Sold by: Weights & Biases
Deployed on AWS
Weights & Biases provides AI developers with the tools needed to build models faster, fine-tune LLMs, and develop GenAI applications with confidence for enterprises of all sizes in any vertical.
4.4
Overview
Weights & Biases provides AI developers with the tools needed to build models faster, fine-tune LLMs, and develop GenAI applications with confidence for enterprises of all sizes in any vertical. The company is trusted by over 1,300 customers including more than 30 foundation model builders.
We provide a comprehensive developer platform to productionize AI. W&B Weave helps developers evaluate, monitor, and iterate to deliver LLM-powered applications, and W&B Models enables ML engineers to train, fine-tune, and manage AI models. Weights & Biases brings together all the developer tools you need for AI into a single, unified platform, delivering enterprise-level performance, scaling, governance, and security.
Weights & Biases helps AI teams of all sizes:
- Build system of record for AI
- Run rigorous evaluations of AI applications
- Debug AI applications pre-production and monitor them in production
- Track experiments for reproducibility and governance
- Track lineage for datasets, models, and metadata
- Collect human feedback and annotations
- Create training datasets leveraging production traces
- Share insights interactively with collaborators
- Implement CI/CD for AI models
Highlights
- W&B was created by AI engineers for AI engineers. Our mission is to build the best tools for Artificial Intelligence.
- Weights & Biases is trusted by more than 1M AI practitioners and used by AI leaders including at OpenAI, Cohere, Toyota Research Institute, and others across industries.
- Weights & Biases works seamlessly with any AI framework or existing architecture, whether in the cloud or on your own infrastructure.
Details
Sold by
Categories
Delivery method
Deployed on AWS
New
Introducing multi-product solutions
You can now purchase comprehensive solutions tailored to use cases and industries.
Features and programs
Buyer guide
Gain valuable insights from real users who purchased this product, powered by PeerSpot.
Financing for AWS Marketplace purchases
AWS Marketplace now accepts line of credit payments through the PNC Vendor Finance program. This program is available to select AWS customers in the US, excluding NV, NC, ND, TN, & VT.
Pricing
Pricing is based on the duration and terms of your contract with the vendor, and additional usage. You pay upfront or in installments according to your contract terms with the vendor. This entitles you to a specified quantity of use for the contract duration. Usage-based pricing is in effect for overages or additional usage not covered in the contract. These charges are applied on top of the contract price. If you choose not to renew or replace your contract before the contract end date, access to your entitlements will expire.
Additional AWS infrastructure costs may apply. Use the AWS Pricing Calculator to estimate your infrastructure costs.
Dimension | Description | Cost/12 months |
|---|---|---|
Annual Single User License for W&B Models | Single user license for 12 months of W&B Models | $4,800.00 |
Annual Commitment for W&B Weave, 10GB | Pricing is dependent on estimated usage of the platform. | $25,000.00 |
Dimension | Description | Cost/unit |
|---|---|---|
overage | Storage overage | $0.001 |
Dimensions summary
This listing bills through an annual contract with three separate dimensions you combine based on need. You buy Annual Single User Licenses for W&B Models, priced per user for a 12-month term, to train and manage AI models. You add an Annual Commitment for W&B Weave in 10GB units to develop GenAI applications, with pricing tied to your estimated platform usage. Storage overage covers usage beyond your committed amount and is billed on consumption. Model seats prorate if added mid-term. Storage and Weave usage are measured over time and billed based on what you use.
Top-of-mind questions for buyers
Ingested bytes are the data received, processed, and stored on your behalf. This includes trace metadata and model inputs and outputs you log to Weave. It excludes communication overhead like HTTP headers. Bytes count as ingested once, at the time they are received and stored.
Storage includes artifacts and data logged to runs. W&B measures your usage over the last 30 days using GB-days, then divides by the number of days to get an average. Your usage is rounded to the nearest megabyte at the end of the period.
Model seat licenses are invoiced annually upfront. Model seats prorate if added mid-term, though removed seats do not earn credits. Weave data ingestion and storage overage are billed monthly in arrears based on the usage you actually consume during each month.
wandb.ai
Helpful?
Vendor refund policy
Non-Refundable. Unless otherwise expressly provided for in this agreement or the applicable Order Form, (i) all fees are based on services purchased and not on actual use; and (ii) all fees paid under this agreement are non-refundable.
Legal
Vendor terms and conditions
Upon subscribing to this product, you must acknowledge and agree to the terms and conditions outlined in the vendor's End User License Agreement (EULA) .
Content disclaimer
Vendors are responsible for their product descriptions and other product content. AWS does not warrant that vendors' product descriptions or other product content are accurate, complete, reliable, current, or error-free.
Delivery details
Software as a Service (SaaS)
SaaS delivers cloud-based software applications directly to customers over the internet. You can access these applications through a subscription model. You will pay recurring monthly usage fees through your AWS bill, while AWS handles deployment and infrastructure management, ensuring scalability, reliability, and seamless integration with other AWS services.
Resources
Support
Vendor support
AWS infrastructure support
AWS Support is a one-on-one, fast-response support channel that is staffed 24x7x365 with experienced and technical support engineers. The service helps customers of all sizes and technical abilities to successfully utilize the products and features provided by Amazon Web Services.
Updated weekly
By Weights & Biases
By Scale AI
By Arize AI
Sentiment is AI generated from actual customer reviews on AWS and G2
Functionality
Ease of use
Customer service
Cost effectiveness
Positive reviews
Mixed reviews
Negative reviews
AI generated from product descriptions
Experiment Tracking and Reproducibility
Track experiments with lineage for datasets, models, and metadata to enable reproducibility and governance of AI development workflows.
LLM Fine-tuning and Model Management
Fine-tune large language models and manage AI models through integrated tools for training, versioning, and lifecycle management.
LLM Application Evaluation and Monitoring
Evaluate, monitor, and iterate on LLM-powered applications with tools for pre-production debugging and production monitoring.
Framework Agnostic Integration
Support for seamless integration with any AI framework or existing architecture, deployable in cloud or on-premises infrastructure.
AI Governance and Lineage Tracking
Implement governance controls with comprehensive tracking of datasets, models, and metadata lineage, including human feedback collection and CI/CD for AI models.
Model Performance Evaluation
Human and machine-based evaluations leveraging AWS Bedrock to assess GenAI application performance, with options for subject matter expert evaluation or automated assessment methodologies.
Industry Benchmarking
Curated industry benchmarks enabling comparison of GenAI applications against industry peers and use cases with regularly refreshed standards.
Vulnerability Assessment
Red teaming capabilities to identify and assess security vulnerabilities and potential failure modes in GenAI applications.
Data Preparation and Optimization
Data processing capabilities including chunking, embedding generation, and RAG knowledge base construction for improved retrieval performance.
Flexible Deployment Architecture
Deployment options supporting both SaaS-based and customer-hosted AWS VPC deployment models.
Agent and Application Observability
Full visibility into AI agent behavior through tree-structured traces capturing user inputs, routing logic, tool calls, memory access, and model outputs with native support for Amazon Bedrock Agents and open-source frameworks
Prompt Optimization and Testing
Prompt IDE environment enabling design, testing, and comparison of prompt versions with live inputs, outputs, and integrated evaluation results for iterative improvement
LLM and Agent Evaluation
Offline and online LLM-as-a-Judge evaluations assessing accuracy, tool-calling, planning, and goal achievement across agent workflows
Closed-Loop Improvement Workflows
Self-improving agent capabilities combining trace analysis, evaluation feedback, and golden datasets for continuous iteration and performance enhancement
Real-Time Monitoring and Alerting
Custom metrics definition and monitoring of latency, token usage, and failures with alert configuration for production issue detection and prevention
Standard contract
No
No
No
Customer reviews
Prerna T.
Makes Comparing AI Experiments Easy, with Everything in One Place
Reviewed on Aug 27, 2026
Review provided by G2
What do you like best about the product?
It makes it much easier to compare different AI experiments i performed. I don't have to manually keep track of different results or parameter, so it helps everything to simply be in one place.
What do you dislike about the product?
I dislike that at sometimes it can get very overwhelming when we go towards advanced features like custom dashboards, Artifacts etc.
What problems is the product solving and how is that benefiting you?
It helps me in keeping track of different AI/ML experiments i performed. For me it is not possible to always remember which model or experiment worked the best and what were the specific results. So, it helps me in that a lot.
Ayush A.
Scalable and Accessible, but a Cluttered UI and Pushy Pricing
Reviewed on Aug 27, 2026
Review provided by G2
What do you like best about the product?
Democratization, ergo scalability. Whether you use it as a student or as a professional, it can become a standard platform—if you know what I mean.
What do you dislike about the product?
It’s a cluttered UI, and the pricing makes it feel like they’re afraid you’ll leave, so they try to squeeze more out of you.
What problems is the product solving and how is that benefiting you?
It helps reduce audit risk by providing complete data on what was done, such as which experiments were run. This makes it easier to see whether employees are wasting resources.
and learning too - older runs can teach new staff
and learning too - older runs can teach new staff
Anson T.
Seamless ML Experiment Tracking with a Clean UI and Effortless Integrations
Reviewed on Aug 25, 2026
Review provided by G2
What do you like best about the product?
What stands out most about Weights & Biases is how seamlessly it tracks ML experiments through a clean, intuitive UI/UX. It integrates effortlessly with frameworks like PyTorch and Hugging Face, which makes real-time performance monitoring and fast, reliable data logging feel almost automatic. Onboarding is quick and well supported by strong documentation, so it’s easy to get up and running without friction. The AI intelligence features also help keep artifact management organized and make hyperparameter evaluation straightforward, which contributes to a fantastic overall ROI.
What do you dislike about the product?
The main drawback of Weights & Biases is that data and storage costs can ramp up quickly when you’re logging very large runs, high-resolution artifacts, or big media files. Also, although the UI is packed with features, the sheer number of metrics and customizable panels can feel overwhelming at first and lead to a steep learning curve for new team members during onboarding.
What problems is the product solving and how is that benefiting you?
Weights & Biases addresses the problem of fragmented machine learning workflows by offering a centralized platform to track experiments, version models, and visualize metrics in one place. It removes the need for manual logging and for keeping results scattered across notebooks. For me, this saves a lot of time during model evaluation, makes it much easier to debug hyperparameter performance, and helps ensure full reproducibility across complex training runs.
Sangeeta S.
Helps track ML experiments.
Reviewed on Aug 24, 2026
Review provided by G2
What do you like best about the product?
It is great that W&B allows me to track various machine learning experiments in one place. When training models like XGBoost, I can compare runs, parameters, and evaluation metrics without having to track everything separately. The dashboard also makes it easy to see which model is training better.
What do you dislike about the product?
The UI might seem a bit complex at first because of the sheer number of features. For smaller projects, it offers more features than I actually need.
What problems is the product solving and how is that benefiting you?
It helps me keep track of my model experiments so I don't forget which versions I tried, which parameters I used, and what results I obtained. Before using W&B, comparing model runs required digging through each individual run, which was very time-consuming. Now, I can quickly review the history and results of past experiments and make informed decisions while improving my models.
Arpit C.
Effortless MLOps and Experiment Tracking with Powerful Real-Time Dashboards
Reviewed on Aug 23, 2026
Review provided by G2
What do you like best about the product?
Weights & Biases makes MLOps and experiment tracking feel effortless, with only minimal code required to get started (wandb.init()). Its real-time, interactive dashboards deliver immediate and detailed visualizations of training loss curves, the impact of hyperparameters, and system resource utilization (GPU/CPU). On top of that, robust artifact versioning and seamless integrations with frameworks like PyTorch, Hugging Face, and TensorFlow make model reproducibility and team-wide collaboration especially smooth. Our experience with Weights & Biases’ automated features has been outstanding, particularly in reducing manual trial-and-error during model training. Features like automated hyperparameter sweeps, background system metric logging, and threshold-based alerts remove the operational friction of running large-scale ML jobs. Instead of constantly monitoring training runs, we rely on automated early-stopping rules and webhook notifications to keep our pipeline efficient.
What do you dislike about the product?
While W&B handles standard metric logging smoothly, the web console can suffer from noticeable UI latency when you load projects with hundreds of concurrent runs or with dense media artifacts. On top of that, keeping the workspace tidy can be a bit of a slog—for example, filtering and bulk-deleting failed experimental runs feels tedious. Finally, per-user pricing can scale steeply as you move from a small team workspace to an enterprise-wide deployment.
What problems is the product solving and how is that benefiting you?
Before implementing Weights & Biases, we struggled with fragmented experiment tracking split between local TensorBoard logs and spreadsheet records. That setup made it hard for our engineering team to compare hyperparameter variations consistently and to reproduce model results.
Weights & Biases addressed this by giving us a unified, real-time dashboard with automated logging for loss curves, system metrics, hyperparameter sweeps, and model artifacts. We can now quickly compare dozens of concurrent training runs, set up automated early stopping for underperforming models, and keep strict version control over dataset-to-model lineage. As a result, we’ve cut down on wasted cloud GPU compute and significantly sped up our iteration cycle from model development through production deployment.
Weights & Biases addressed this by giving us a unified, real-time dashboard with automated logging for loss curves, system metrics, hyperparameter sweeps, and model artifacts. We can now quickly compare dozens of concurrent training runs, set up automated early stopping for underperforming models, and keep strict version control over dataset-to-model lineage. As a result, we’ve cut down on wasted cloud GPU compute and significantly sped up our iteration cycle from model development through production deployment.