Both provide AI observability but differ in scope. Arize AX monitors your AI layer, while Logfire covers your entire application in one trace: LLM calls, database queries, APIs, and frontend. When agents fail in production, you see the complete picture, not just what happened inside the model. Arize offers two products: Phoenix, an open source tool for LLM tracing and evaluation, and Arize AX, a commercial SaaS platform for ML and AI monitoring. This page compares Logfire against Arize AX.
| Feature | Logfire | Arize AX |
|---|---|---|
| Observability scope | Full-stack (AI + databases, APIs, infra) | AI/LLM tracing + ML model monitoring |
| Strength | Application tracing + AI | Drift detection, model performance, agent evaluation |
| Non-AI tracing | Full support | OTel-compatible but no first-class support for web frameworks or databases |
| Language support | Python, JS/TS, Rust + any OTel language | Python SDK plus OTel/OpenInference integrations for JS/TS, Java, and Go |
| Pricing | Personal: 10M records free; Team: $49/month + $2/M over 10M | $50 + span and payload usage (AX Pro) |
| Data retention | 30 days default; 90 days (Growth); custom on Enterprise | 30 days on AX Pro; custom on Enterprise |
| Query interface | PostgreSQL-compatible SQL | AQL (SQL-like) plus product filters and Alyx |
| Human annotations | Annotations on production runs | Annotations on production runs |
| MCP server | Query production traces from your editor or AI agent | Instrumentation guidance and docs; production trace querying not available |
| Workload | Arize AX | Logfire | Difference |
|---|---|---|---|
| 1 user, 5M spans/mo | ~$4,055 | ~$49 | ~83x |
| 5 users, 50M spans/mo | ~$40,730 | ~$129 | ~316x |
| 20 users, 500M spans/mo | ~$407,480 | ~$1,229 | ~332x |
Calculation scope
Arize AX Pro span and payload pricing compared with the cheapest applicable public Logfire plan at the same span count and 30-day retention. Payload uses a disclosed 5 KB modeled average.
Basis: USD; public monthly list price. Prices checked .
The Arize payload model uses 5M × 5 KB = 25 GB per month.
$49 $49 (5M − 10M included) ÷ 1M × $2 $0 $50 $50 (5M − 50K included) × $0.0008 $3,960 max(25 GB − 10 GB included, 0) × $3/GB $45 The Arize payload model uses 50M × 5 KB = 250 GB per month.
$49 $49 (50M − 10M included) ÷ 1M × $2 $80 $50 $50 (50M − 50K included) × $0.0008 $39,960 max(250 GB − 10 GB included, 0) × $3/GB $720 The Arize payload model uses 500M × 5 KB = 2,500 GB per month.
$249 $249 (500M − 10M included) ÷ 1M × $2 $980 $50 $50 (500M − 50K included) × $0.0008 $399,960 max(2,500 GB − 10 GB included, 0) × $3/GB $7,470 “Covers 10 million logs, spans, and metrics every month.”
The official calculator shows 20 million records and no extra seats totaling $69/month on Team.
“$.0008 per additional span”
AX Pro lists 50,000 spans and 10 GB included, then $0.0008 per additional span and $3 per additional GB.
When something breaks in production, you need to know whether it was the model, the data, or the infrastructure. Arize AX shows you the AI layer. Logfire shows you the whole thing: the user request, the database query, the LLM call, the API response. One trace, complete context. Arize AX can receive OTel spans from non-AI services, but has no first-class support for web frameworks or databases.
Logfire's Team plan starts at $49 a month and includes 10 million records; additional records cost $2 per million. Within the published 5 KB average span allowance, there is no separate per-GB payload meter. That makes the cost of retaining complete production traces easier to model from record count alone.
Arize custom metrics use AQL, a vendor-specific SQL-like language, and Alyx can generate queries and trace filters in the product. Logfire uses PostgreSQL-compatible SQL across telemetry and exposes the same query workflow through its MCP server. Queries can live in scripts and pull requests, power dashboards and alerts, and be written by coding agents using an interface they already understand.
Arize AX comes from the ML monitoring world, excelling at model drift detection, performance metrics, and agent evaluation. Logfire comes from the application observability world, excelling at distributed tracing across your entire stack, databases, APIs, infrastructure monitoring, and real-time debugging, alongside AI. The biggest practical difference is scope. Logfire sees your whole application. Arize AX focuses on the model and AI layer.
Yes. Arize Phoenix is a separate open-source, self-hosted product. This page compares the managed Logfire Cloud and Arize AX offerings; it does not assign a monthly price to Phoenix or try to estimate organization-specific self-hosting costs.
Dynatrace agreed to acquire Arize in August 2026, aiming to combine Arize's AI evaluation and observability capabilities with Dynatrace's application and infrastructure tracing. The deal is expected to close later in 2026. Pydantic Logfire combines full-stack and AI observability in one product from the start, and hasn't been acquired or merged into another platform. The feature and pricing comparisons on this page reflect Arize AX as it stands today, independent of Dynatrace.
With Logfire, three statements get you started: import logfire, logfire.configure(), logfire.instrument_openai(). From there, FastAPI, SQLAlchemy, HTTPX, and other common application integrations are one call each. Arize AX documents more than 30 AI provider and framework integrations across Python, JavaScript, and Java. For general web-framework, database, and background-worker tracing, you use standard OpenTelemetry instrumentation rather than AX-specific application integrations.
Logfire focuses on application observability and real-time AI tracing. For specialised ML model monitoring features like drift detection and model performance metrics, Arize AX is better suited. You can use both tools together.
Yes. Use Logfire for full-stack application observability, real-time debugging, and querying your traces via SQL or MCP. Use Arize AX for specialised ML monitoring like drift detection and model performance metrics. Both are built on OpenTelemetry, so they integrate cleanly.
Yes. With supported client and server instrumentation in place, OpenTelemetry propagates trace context and Logfire reconstructs spans from different services into one trace. Logfire's integrations handle propagation automatically for common frameworks; custom transports still need explicit context propagation.
Get started with 10 million free spans, logs, and metrics per month. No credit card required.