Positioning
Weave is the LLM evaluation and observability product from Weights and Biases. It sits inside the broader W&B Models stack and reuses W&B's project, run, and dataset primitives. The product covers trace trees, auto-versioning of code and prompts, online eval that fires on production traces, multimodal tracking, and a leaderboards surface. The strongest fit is teams whose ML stack is already on W&B and who want to keep evals in the same console as model training.
Ownership and URLs have moved. CoreWeave completed its acquisition of Weights and Biases on May 5, 2025 1, and Weave's docs and pricing now live under CoreWeave: the old wandb.ai pricing URL redirects to coreweave.com/forge-pricing, the docs are at docs.coreweave.com/weave, and the quickstart asks for a CoreWeave Forge account and API key 2. The docs now lead with agent tracing (OTel-compatible SDKs, with sessions, turns, and sub-agents as first-class concepts). Weave monitors score production traces with built-in quality signals (hallucination, user frustration, jailbreaking, and others) and error signals such as rate limiting. They judge successful root-level traces for quality and failed traces for errors; child spans and intermediate calls are not scored 3.
Strengths
- Native to W&B. If your team already uses W&B for experiment tracking, dataset versioning, and model registry, Weave drops in without a parallel toolchain. Project structure, permissions, and links across runs are reused.
- Multimodal tracking. Image, video, audio, and PDF content are first-class artifacts in the Weave UI, which is useful for vision and multimodal RAG evals where text-only platforms render outputs awkwardly.
- Auto-versioning of code and prompts. Calling a Weave op creates a new op version whenever its code changed since the last call and logs the inputs and outputs, so a regression in eval scores can be traced to a code change without manual bookkeeping 2. Prompts published with
weave.publish are versioned as objects. This kind of provenance is a recurring theme in production LLM platform writeups 4.
Limitations
- Tied to the W&B account model. Teams not already on W&B are paying for both the Models stack and Weave to get the integration benefit. The cost-benefit calculation changes when Weave is the only product you would buy from W&B.
- Less momentum in the eval-only narrative. Compared to Langfuse, Phoenix, or Braintrust, public examples of Weave-first eval programs are thinner. The Weave repo shows 1.1k GitHub stars against 11.7k for Phoenix and 22.3k for Opik (as of 2026-10-01), and the product page names Canva, Microsoft, and Toyota as customers. The product is real but the public story is narrower.
- Vendor-published benchmarks. Performance claims published by W&B are vendor-published; treat them as positioning, not independent measurement.
Best fit
Teams already running the W&B Models stack who want a single console for training, fine-tuning, and eval. Multimodal eval programs where image or video outputs are central. Production observability where the same team owns model training and serving.
Getting started
import weave
from weave import Evaluation
weave.init("rag-prod")
@weave.op # captures inputs, outputs, code version, and parameters
def app(question: str, context: list[str]) -> str:
return run_rag(question, context)
def accuracy(output: str, expected: str) -> dict:
return {"correct": int(expected.lower() in output.lower())}
eval = Evaluation(
dataset=[
{"question": "Is the dish vegan?", "context": ["Lentil bowl: no dairy."], "expected": "yes"},
],
scorers=[accuracy],
)
await eval.evaluate(app)
Trace trees, the auto-recorded code version, and the per-example scores render in the W&B UI. Subsequent runs against the same dataset diff inside Weave. Weave passes the app's return value to scorers as output; the older model_output argument name is legacy 5. This snippet was updated from the current docs and was not run here.
Pricing notes
Pricing is published on the CoreWeave Forge pricing page (fetched as of 2026-10-01) 6. Free is 60 per month with 100 GB of storage, 1.5 GB per month of Weave ingestion, and up to 10 model seats, and is pitched at teams under 50 employees. Additional Weave ingestion on Pro is 100 per extra GB, so estimate your trace payload size (inputs, outputs, and media) before picking a tier. Enterprise is quote-based and adds SSO, audit logs, and a HIPAA option. A free academic plan includes up to 25 GB per month of Weave ingestion. The Weave SDK is Apache 2.0 on GitHub 7; the hosted service is what you pay for. Open-source ML tool surveys consistently call out W&B as a heavyweight in the broader ML observability category 8.
Alternatives
Consider Langfuse or Arize Phoenix for OSS-first observability. Consider Braintrust for a hosted, framework-agnostic console. Consider LangSmith for LangChain-heavy teams. For pure CI regression, DeepEval or Promptfoo.
Citations and last verified
Verified 2026-10-01 against the W&B Weave product page, the Weave docs at docs.coreweave.com, the PyPI listing, the CoreWeave acquisition press release, and the CoreWeave Forge pricing page.