Both trace agents, score production traces with LLM judges and code, run annotation queues and version prompts. Opik is a standalone open-source platform you can run yourself for free; W&B Weave now sits inside CoreWeave Forge and fits teams that already use Weights & Biases for model work. Where your training stack lives often decides it.[1][6]
Opik and W&B Weave both trace and evaluate LLM agents; the difference is that Opik's platform is Apache-2.0 and free to self-host, while self-hosting Weave needs a W&B licence and an existing W&B Platform install. Weave belongs to CoreWeave, which completed its acquisition of Weights & Biases in May 2025, and since 30 September 2026 W&B hosted accounts sign in through CoreWeave Forge. Weave suits teams already training models on W&B; Opik suits teams that want a standalone open-source platform.
| Decision factor | Opik | W&B Weave |
|---|---|---|
| Ownership | Comet ML, independent; reports $70M in total funding | CoreWeave (acquired W&B, May 2025); sold through CoreWeave Forge since 30 September 2026 |
| Licence | Apache-2.0 core platform | Weave repository Apache-2.0 |
| Self-hosting | Free; Kubernetes for production; user administration excluded | Kubernetes with ClickHouse; needs a Weave-enabled W&B licence and W&B Platform install |
| SDKs | Python, TypeScript, Java, .NET; Ruby via OpenTelemetry | Python and TypeScript; TypeScript lacks guardrails, class-based scorers and cost tracking |
| Online evals | Rules score sampled traces, spans and threads | Monitors with LLM judges; docs now call these a 'previous approach' |
| Guardrails | Called from your code; fail closed with an exception | Inline scorers called from Python code; can block or modify inputs and outputs |
| Free tier | Free Cloud: 25k spans a month, 60-day retention | Forge Free: 1 GB of Weave ingestion a month |
| First paid tier | Pro Cloud: $19 a month, 100k spans | Forge Pro: from $60 a month, 1.5 GB of Weave ingestion; teams under 50 employees |
| Capability | Opik | W&B Weave | Prefactor |
|---|---|---|---|
| Observe | |||
| LLM call tracing | ✓ | ✓ | ✓ |
| Multi-step agent tracing (tool calls, nested spans) | ✓ | ✓ | ✓ |
| Cost tracking | ✓ | ✓ | ✓ |
| OpenTelemetry support | ✓ | ✓ | ✓ |
| Evaluate | |||
| Offline evals (datasets, experiments) | ✓ | ✓ | ✓ |
| Online evals on production traffic | ✓ | ✓ | ✓ |
| LLM-as-a-judge scoring | ✓ | ✓ | ✓ |
| Code-based evaluators | ✓ | ✓ | ✓ |
| Human review and annotation | ✓ | ✓ | ✓ |
| Prompt versioning and playground | ✓ | ✓ | — |
| Act on findings | |||
| Runtime controls on live agent actions | Partial | Partial | ✓ |
| Gateway or proxy in the request path | Partial | Not stated | — |
| Licence and deployment | |||
| Open source | ✓ | Partial | Partial |
| Self-hosting | ✓ | ✓ | Partial |
CoreWeave completed its acquisition of Weights & Biases on 5 May 2025. On 30 September 2026 it launched CoreWeave Forge, which bundles W&B Models, Weave and new CoreWeave tools; W&B hosted accounts now sign in through Forge, while self-hosted and dedicated-instance accounts are unchanged. Weave's pricing, documentation and login now redirect to coreweave.com domains.
Within Weave, the direction for agents is shifting too. CoreWeave now recommends Agent Lens, a separate product in public preview that shares Weave's trace backend, for agent tracing and analysis, and Weave's documentation describes its built-in monitors as a 'previous approach'. If you are choosing today, expect packaging to keep moving.
Opik is built by Comet ML, which is independent. Its 2026 changelog shows steady releases, including Opik 2.0 with an environment-tagged Prompt Library in June and automatic annotation queues in October.
Agent Lens has its own SDK and UI, so a team adopting Weave for agents today may face a second move later.
Both repositories are Apache-2.0, but the terms for running them differ. Opik's self-hosted version includes every feature except user administration and needs no licence; Kubernetes is the documented production route.
Self-hosting Weave requires a Weave-enabled licence from W&B and an existing W&B Platform installation, on Kubernetes 1.29 or later with ClickHouse and S3-compatible storage. Dedicated Cloud is the single-tenant alternative. For a team without W&B already, Weave's self-hosted path is a larger commitment than Opik's.
At the Enterprise level, Opik adds unlimited users, SSO and on-premises or fully hosted deployment. Weave's Enterprise quote adds SSO, SCIM, customer-held keys, private connectivity and bring-your-own-bucket.
Weave's Python SDK is the complete one. The TypeScript SDK has no class-based scorers, no guardrails and no cost tracking. Weave accepts OTLP over HTTP and understands OTel GenAI, OpenInference and OpenLLMetry conventions, and since June 2026 its OpenTelemetry integrations are on by default. Its strongest fit is a team that already tracks model training in W&B and wants agent traces, evaluations and a Registry in the same place.
Opik's SDKs cover Python, TypeScript, Java and .NET, and it lists more than 30 framework integrations. It accepts OTLP over HTTP only. Its guardrails, like Weave's, are checks your code calls, and PII and topic checks are English only. Neither product documents approvals on agent tool actions.
Evaluation features are broadly matched. Weave's Evaluation API runs a model over a dataset with scorers and shares results as leaderboards; it has LLM-as-judge scorers, local small-model scorers for toxicity and context relevance, and annotation queues that export to datasets. Opik offers datasets, experiments and Test Suites, more than 30 built-in metrics, an 'Agent as a Judge' and a Prompt Optimizer with six algorithms. Weave calculates cost with cache tokens for supported providers, in Python only; Opik's Cost Intelligence extends cost tracking to coding agents such as Claude Code and Codex.
As of 20 October 2026, Opik prices by spans: self-hosting is free with unlimited spans, Free Cloud covers 25k spans a month, and Pro Cloud is $19 a month for 100k spans, then $5 per 100k.
Weave is priced as part of CoreWeave Forge, by data ingested. Forge Free includes 1 GB of Weave ingestion a month. Forge Pro starts at $60 a month with 1.5 GB of Weave ingestion, up to 10 model seats and 100 GB of storage, for teams under 50 employees; extra ingestion is listed at $0.10 per MB. Large traces with long prompts use ingestion faster than span counts suggest, so measure your average trace size against Opik's pricing and CoreWeave Forge pricing.
Forge Pro also bundles seats and storage for W&B Models, which matters only if you use them. Enterprise ingestion overage is listed at $0.20 per MB.
We sell the layer this section describes. Read it with that in mind.
Stated plainly, since Prefactor publishes this page: Prefactor is an agent evaluation platform that scores each step of a run and acts on what it finds through runtime policies on tool calls and data access, approval holds for risky actions, and a kill switch for a run, an agent or a team. Opik and Weave both offer guardrails that your code calls to check inputs and outputs; neither holds an agent's tool action for approval. Prefactor has no prompt playground or prompt registry and its platform is hosted, so for prompt iteration, free self-hosting or a link to W&B model training, Opik or Weave is the better pick. See the full comparison of 15 platforms.
Reviewed against public product and documentation pages on October 20, 2026. If a vendor has changed a feature, product name, or positioning since then, send a correction and we will update it. Numbered source links in the page body point to the ordered sources below.
Prefactor scores each step of every agent run and can hold or stop an action when a score says it should not happen. Start on the free tier: 5,000 spans a month, no credit card required.
Prefactor helps teams observe, evaluate, and improve their AI agents in production, across every framework and provider.