Both are built on OpenTelemetry and both score production traffic. Arize gives you a free self-hosted starting point in Phoenix and a paid path to AX, now under Dynatrace ownership; Logfire puts agent traces next to your HTTP, database and log data and adds an AI gateway that can block model calls. Whether you need to run the software yourself decides most of this choice.[3][4]
Arize is an evaluation and observability vendor with a free self-hosted tool (Phoenix) and a paid platform (AX); Logfire is Pydantic's hosted OpenTelemetry platform that traces the whole application, not only model calls. Dynatrace completed its acquisition of Arize on 1 October 2026, and Pydantic, which makes Logfire, is independent. Phoenix can be self-hosted for free, while self-hosted Logfire is Enterprise only.
| Decision factor | Arize | Logfire |
|---|---|---|
| Ownership | Acquired by Dynatrace, completed 1 October 2026; Phoenix and AX continue | Pydantic, independent; $12.5M Series A led by Sequoia in 2024 |
| Trace scope | LLM and agent applications | Whole application: HTTP, database, logs and metrics, plus LLM and agent views |
| Licence | Phoenix: Elastic License 2.0 (source-available); AX: commercial | SDKs and pydantic-evals MIT; platform closed source |
| Self-hosting | Phoenix free; AX on Enterprise, including air-gapped | Enterprise only, on Kubernetes via Helm |
| Production evals | AX only: evaluator tasks run on stored production traces | Evaluators run in your application in the background |
| Run-time checks | Partial: AX guards run in your code via Guardrails AI | Partial: gateway guardrails and spend caps on model calls |
| Free tier | AX Free: 25k spans a month, 15-day retention | Personal: 10M records a month, 1 seat, 30-day retention |
| First paid tier | AX Pro: $50 a month, 50k spans | Team: $49 a month, 10M records, 5 seats |
| Capability | Arize | Logfire | 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 | ✓ | Partial | ✓ |
| Prompt versioning and playground | ✓ | ✓ | — |
| Act on findings | |||
| Runtime controls on live agent actions | Partial | Partial | ✓ |
| Gateway or proxy in the request path | Partial | ✓ | — |
| Licence and deployment | |||
| Open source | Partial | Partial | Partial |
| Self-hosting | ✓ | Partial | Partial |
Dynatrace completed its acquisition of Arize on 1 October 2026. Its announcement says Phoenix and AX will both continue and that Arize capabilities will be integrated into Dynatrace over time, so packaging may change. Pydantic, the company behind the Pydantic library, Pydantic AI and Logfire, is independent; its last confirmed round was a $12.5M Series A led by Sequoia in October 2024.
The bigger practical difference is what you can run on your own servers without a contract. Phoenix installs with pip, Docker or Helm and costs nothing to self-host. Its code is public under the Elastic License 2.0, which allows self-hosting and modification but forbids offering it to others as a hosted service. Logfire's platform is closed source. Its SDKs and the pydantic-evals library are MIT-licensed, but self-hosting the platform needs the Enterprise plan, a Kubernetes cluster, PostgreSQL 16 or later and object storage. If traces must stay in your network and there is no budget yet, Phoenix is the option that works today.
Arize focuses on LLM and agent applications. It maintains OpenInference, a set of OpenTelemetry conventions for model and agent spans, and AX draws an Agent Trajectory graph of which agents call which, set automatically for LangGraph, CrewAI and AG2. Phoenix added decision spans for routers and classifiers in October 2026.
Logfire is a general observability platform with dedicated Agents and LLMs views. Its SDKs are OpenTelemetry wrappers, so one trace can show an incoming HTTP request, the database queries and the model and tool calls an agent made in between, alongside logs and metrics. Integrations cover Pydantic AI, LangGraph, CrewAI, Google ADK, Mastra and the Vercel AI SDK among others, and any standard OpenTelemetry SDK can send data, though the richest tooling is Python-first. If agent failures in your system are often tangled with slow queries or upstream errors, one trace for both helps. If you already have an APM tool for that, Arize's narrower focus costs you little.
Both run offline experiments on datasets with LLM judges and code scorers. They differ on scoring live traffic. In AX, evaluator tasks run continuously against stored production traces at span, trace or session level, with sampling and filters; Phoenix's documentation points to AX for this. In Logfire, live evaluations run inside your application through pydantic-evals, in the background at a sample rate you set, and emit OpenTelemetry events that Logfire aggregates. Evaluator code stays in your repository, and it also runs on your compute.
Human review is further along at Arize: Phoenix and AX both support annotations, and AX has annotation queues that accept spans, traces, sessions or dataset examples. Logfire's run annotations are in beta and its annotation queues are an early-access feature.
Neither covers agent tool actions at run time. AX documents guards built on Guardrails AI that run in your own code around model calls. Logfire's AI gateway, folded into Logfire in March 2026, can observe, flag, redact or block model calls containing secrets or personal data and stop requests once a spend cap is reached, but only for traffic routed through it.
As of 20 October 2026, AX prices on spans: Free covers 25k spans a month with 15-day retention, and Pro is $50 a month for 50k spans with 30-day retention, both with unlimited users. Logfire prices on records, which include spans, logs and metrics: Personal is free for 10M records a month with one seat, Team is $49 a month with 10M records and five seats, and Growth is $249 a month with unlimited seats. Extra records cost $2 per million on paid plans.
Logfire counts every span in the application, not only model calls, so estimate your full record volume before comparing. Phoenix remains free if you run it. Check Arize's pricing and Logfire's pricing before you decide.
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. Arize AX's guards run around model calls in your own code, and Logfire's gateway checks model traffic routed through it; neither documents approvals or policies on agent tool calls. Prefactor has no prompt playground or prompt registry and its platform is hosted, so if free self-hosting is the requirement, Phoenix 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.