Both trace multi-step agents, run evaluations on datasets, score production traffic and version prompts, and both now belong to larger infrastructure companies. Arize suits teams that want a free self-hosted start and OpenTelemetry at the centre of their instrumentation; Weave suits teams that already train models in W&B or run on CoreWeave.[3][6]
The difference between Arize and W&B Weave is where each one sits: Arize is a standalone evaluation vendor with a free self-hosted tool (Phoenix) and a hosted platform (AX), while Weave is the LLM and agent part of Weights & Biases, now sold inside CoreWeave Forge. Dynatrace completed its acquisition of Arize on 1 October 2026; CoreWeave bought W&B in May 2025 and moved W&B hosted accounts to Forge in September 2026. Phoenix self-hosts for free, while self-hosting Weave needs a W&B licence and a W&B Platform install.
| Decision factor | Arize | W&B Weave |
|---|---|---|
| Ownership | Acquired by Dynatrace, completed 1 October 2026; Phoenix and AX continue | CoreWeave since May 2025; sold inside CoreWeave Forge since 30 September 2026 |
| Products | Phoenix (self-hosted) and Arize AX (hosted, or self-hosted on Enterprise) | Weave, plus Agent Lens, a separate agent product in public preview |
| Licence | Phoenix: Elastic License 2.0 (source-available); AX: commercial | Repository is Apache-2.0; self-hosting needs a Weave-enabled licence from W&B |
| Self-hosting | Phoenix free; AX on Enterprise, including air-gapped | Kubernetes and ClickHouse, on top of a W&B Platform install |
| Online evals | AX only; Phoenix docs point to AX for continuous production evals | Monitors score production traces; docs now call these a previous approach and recommend Signals |
| Runtime guardrails | Partial: AX guards run in your code via Guardrails AI | Partial: scorers called from Python code can block or modify inputs and outputs |
| Free tier | AX Free: 25k spans a month, 15-day retention | Forge Free: 1 GB of Weave ingestion a month, up to 5 model seats |
| First paid tier | AX Pro: $50 a month, 50k spans | Forge Pro: from $60 a month, 1.5 GB of Weave ingestion |
| Capability | Arize | 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 | Partial |
| Self-hosting | ✓ | ✓ | Partial |
Arize's change of owner is the more recent. Dynatrace completed its acquisition on 1 October 2026 and says both Phoenix and AX will continue, with Arize capabilities integrated into Dynatrace over time. That integration has not happened yet, so the products you evaluate today are the ones Arize built.
Weave's change is further along. CoreWeave completed its acquisition of Weights & Biases in May 2025, and 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, and Weave's pricing, documentation and login have moved to CoreWeave domains. The same launch put Agent Lens, a separate agent observability product that shares Weave's trace backend, into public preview, and CoreWeave now recommends it for agent tracing. Expect Weave's packaging to keep moving while that settles.
For a team that wants to start without a contract, this is the clearest difference. Phoenix installs with pip, Docker or a Helm chart and costs nothing. It covers tracing, datasets, experiments, LLM and code evaluators and prompt versioning. Continuous evaluation of production traffic, monitors and alerting are AX features, and Phoenix's documentation points you there. Phoenix uses the Elastic License 2.0, which lets you modify and self-host it but not offer it to others as a hosted service.
Weave's code is in an Apache-2.0 repository, but a self-hosted deployment needs a Weave-enabled licence from W&B and an existing W&B Platform installation, running on Kubernetes 1.29 or later with ClickHouse and S3-compatible storage. Dedicated single-tenant cloud is the other option. In practice, self-hosting Weave is a commercial conversation, and it makes most sense for teams that already self-host W&B for model training. Arize AX can also run on your own Kubernetes, including air-gapped, but only on the Enterprise plan.
Both accept OpenTelemetry. Arize maintains OpenInference, the span conventions for LLM and agent work that several other tools read, and Phoenix is built on OpenTelemetry. Weave accepts OTLP over HTTP and understands OTel GenAI, OpenInference and OpenLLMetry attributes; since its June 2026 SDK release its OpenTelemetry integrations are on by default. Instrumentation written for one can usually be pointed at the other.
Weave's own SDK centres on a Python decorator, and its TypeScript SDK does less: no class-based scorers, guardrails or cost tracking. Arize offers Python and TypeScript SDKs plus a Go client for Phoenix. For multi-agent work, AX draws an agent trajectory graph set automatically for frameworks such as LangGraph and CrewAI; Weave groups calls into threads and shows OTel GenAI spans in an Agents view.
Neither platform intercepts traffic on its own. AX documents guards built on Guardrails AI that run in your code around model calls. Weave's guardrails are scorers you call from Python application code that can block or modify inputs and outputs. Neither documents approvals or policies on an agent's tool calls.
The two count different things. As of 20 October 2026, Arize AX prices by spans: AX Free covers 25k spans a month with 1 GB of ingestion and 15-day retention, and AX Pro is $50 a month for 50k spans, 10 GB of ingestion and 30-day retention, both with unlimited users. Weave is priced through CoreWeave Forge by seats, storage and Weave ingestion: Forge Free includes up to 5 model seats and 1 GB of Weave ingestion a month, and Forge Pro starts at $60 a month with up to 10 model seats and 1.5 GB, for teams under 50 employees. The pricing page lists extra Weave ingestion at $0.10 per MB.
Span counts favour teams with large spans; byte counts favour teams with many small ones. Forge tiers also cover W&B model tracking, which matters only if you use it. Check Arize's pricing and CoreWeave Forge pricing against your own trace volumes.
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, not only the final answer, 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. Arize and Weave both offer partial run-time checks, guards or scorers you call from your own code around model inputs and outputs, and neither documents approvals on agent actions. Prefactor has no prompt playground or prompt registry and its platform is hosted, so if free self-hosting or prompt iteration is the requirement, Phoenix or Weave is the better fit. 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.