Weights & Biases

The AI developer platform for experiment tracking and LLMOps.

Hand-reviewed by the Toolglade team·Updated Aug 2026

coding#mlops#llmops#experiment-tracking#llm-observability

Free plan Claimed API Self-hosted Teams

Toolglade’s take

Weights & Biases is a mature, widely adopted platform that now covers both classic ML experiment tracking (Models) and LLM observability and evaluation (Weave). The free tier is genuinely useful for individuals and research, and Weave's one-decorator tracing is low-friction. Watch the non-commercial restriction on the free tier, the new 2026 usage-metered billing that makes costs harder to predict, and CoreWeave ownership, which raises questions about infrastructure neutrality and lock-in.

About Weights & Biases

Weights & Biases is a mature AI developer platform combining ML experiment tracking, model versioning, and sweeps (W&B Models) with LLM observability and evaluation (W&B Weave). Weave adds automatic tracing via a decorator, evaluation pipelines with human and automated scoring, and guardrails. It is now a CoreWeave business unit after a roughly $1.7B acquisition, with a free tier for individuals and usage-metered paid plans.

Weights & Biases became an industry standard for machine learning experiment tracking, offering dataset and model versioning, hyperparameter sweeps, a model registry, and reports. Its W&B Models product remains the go-to for training and fine-tuning workflows, used widely across research and enterprise ML teams. W&B Weave extends the platform into LLMOps, adding automatic tracing of every LLM call, evaluation pipelines with human and automated scoring, dataset versioning, and guardrails including pre-built scorers for toxicity, bias, PII, and hallucination. Together, Models and Weave cover both classic ML and modern LLM app development in one ecosystem. In 2025, CoreWeave acquired W&B for approximately $1.7B, and it now operates as a CoreWeave business unit under its own brand. The deal ties W&B into CoreWeave's GPU infrastructure, including serverless inference and sandboxes. A 2026 pricing restructure introduced usage-metered add-ons for storage, Weave data ingestion, and per-token inference alongside the free, Pro, and Enterprise tiers.

TL;DR

Weights & Biases is a mature AI developer platform spanning ML experiment tracking, versioning, and sweeps (Models) plus LLM observability and evaluation (Weave). Weave offers low-friction tracing via a decorator, evaluation pipelines, and guardrails. It is now a CoreWeave business unit after a roughly $1.7B acquisition completed in 2025. A free tier serves individuals and research, while paid plans start around $50 per user per month with new usage-metered add-ons.

Company overview

Weights & Biases built one of the most widely used ML experiment-tracking platforms, serving more than 1,400 customer organizations and dozens of foundation-model builders. It later added the Weave LLMOps product to cover modern LLM and agent workflows.

CoreWeave acquired W&B for approximately $1.7B, completing the deal in May 2025. W&B now operates as a CoreWeave business unit, tying it into CoreWeave's GPU infrastructure, serverless inference, and sandboxes.

Product features

W&B Models provides experiment tracking, hyperparameter sweeps, artifact and model versioning, a model registry, and reports. W&B Weave adds LLM tracing, evaluation pipelines with pre-built scorers for toxicity, bias, PII, and hallucination, quality scoring, guardrails, and cost, latency, and token tracking.

It integrates with OpenAI, Anthropic, Google Gemini, LangChain, LlamaIndex, and major ML frameworks including PyTorch, TensorFlow, and Hugging Face, and now connects to CoreWeave infrastructure.

Target market

ML and AI engineering and research teams, from individual researchers on the free tier to large enterprises and foundation-model labs. Models suits classic ML training and fine-tuning; Weave suits teams shipping LLM and agent apps needing observability and evaluation.

Buyer personas

End users

ML engineers, researchers, and AI engineers who track experiments, version models, and trace and evaluate LLM apps.

Buyers

ML platform leaders and engineering directors standardizing tooling across teams.

Key influencers

Research leads, MLOps engineers, and data-governance stakeholders.

Ideal customer profile

An ML or AI team wanting a mature, all-in-one platform for both training experiment tracking and production LLM observability, comfortable in a SaaS-first ecosystem.

Funding & performance

Before its acquisition, W&B had raised more than $250M across multiple rounds, including a Series C led by Insight Partners, at a last private valuation of roughly $1.25B. CoreWeave then acquired the company for approximately $1.7B, completing the deal in May 2025. W&B now operates as a wholly owned CoreWeave business unit.

Pros & cons

Pros

Industry-standard, mature experiment tracking with a large user base
Weave gives low-friction LLM tracing with one decorator plus rich evals
Covers both classic ML and modern LLMOps in one ecosystem
Pre-built safety and quality scorers and guardrails out of the box
Genuinely useful free tier for individuals and a strong academic program
Broad framework and LLM-provider integrations
Backed by CoreWeave with deep GPU and inference resources

Cons

Non-commercial restriction on the free tier limits business use
New 2026 usage-metered billing makes costs harder to predict
A team-size threshold pushes growing teams into custom Enterprise quickly
Enterprise seat pricing is high per third-party estimates
Only the Weave SDK is open source; the backend is proprietary and SaaS-first
CoreWeave ownership raises neutrality and lock-in questions
Two overlapping product lines can confuse newcomers

Pricing plans

Personal

Non-commercial use
Unlimited experiments
~100-200GB storage
Community support

Pro

~$50 / month

Per-user pricing
Team collaboration
Email and chat support
Weave observability and evals

Enterprise

Custom

SSO and RBAC
Private, VPC, or on-prem deployment
Compliance
Usage-metered add-ons
Dedicated support

Key features

API

Team collaboration

Self-hosted

Integrations

OpenAI, Anthropic, Google Gemini, LangChain, LlamaIndex, PyTorch, TensorFlow, Hugging Face

Input types

text

Output types

text

Best For

ML experiment tracking, LLM tracing and evaluation, model versioning and registry, research and enterprise ML teams

Compare key features

View all alternatives →

Feature

Weights & Biases

Langfuse

Vellum

Pricing

Freemium

Free plan

Yes

Free trial

Yes

API

Yes

Self-hosted

Yes

Team support

Yes

Frequently asked questions

Is the W&B free tier usable for commercial work?+

No. The free Personal tier is for non-commercial use. Business use requires a Pro or Enterprise plan.

What is the difference between W&B Models and Weave?+

Models is the original ML experiment tracking, versioning, sweeps, and registry for training and fine-tuning. Weave is the LLMOps layer for tracing, evaluation, and guardrails of LLM and agent apps.

Is Weights & Biases open source?+

Only partially. The Weave SDK is open source on GitHub, but the hosted backend and UI are proprietary. Self-managed and on-prem deployment is available for Enterprise.

Who owns Weights & Biases now?+

CoreWeave acquired W&B for approximately $1.7B, with the deal completed in May 2025. W&B operates as a CoreWeave business unit under its own brand.

How does Weave tracing work?+

You add the @weave.op decorator to your functions, and Weave automatically traces each LLM call including inputs, outputs, latency, and cost, which you can then evaluate and analyze.