ComparisonVERIFIED 2026-08-08
SouthBase vs Arize AX
The deepest ML heritage in the category — drift detection, embedding analysis, and serious enterprise compliance. They also author and maintain OpenInference, the span convention we ourselves build on. Phoenix is their open-source, self-hostable tracing tool.
Where Arize AX is better3 CONCESSIONS
Start with what they do well
They own the standard we use
OpenInference is Arize's work, and our span spec is an extension of it. We benefit from their standards effort, and it would be dishonest to present that as anything else.
ML drift and embedding analysis are theirs
We do not compete on drift detection or embedding drilldowns, and we do not intend to. If that is your core problem, Arize is the better product.
Their enterprise compliance posture is ahead of ours
They have the certifications and the enterprise track record. We are pre-launch.
Where we think they fall short
Our actual criticisms
Teams who outgrow Phoenix and discover that moving to AX is a repricing event, and teams whose RAG or agent payloads make GB-metered pricing unpredictable.
- 01
Two metered axes make the bill hard to predict
AX Pro meters spans and gigabytes separately — $0.0008 per extra span and $3 per extra GB. Agent and RAG workloads are exactly the shape that blows through the GB axis first, because retrieved context and full prompt/completion text are large.
- 02
GB metering quietly discourages logging the thing you need
This is the part we object to most. When storage is billed per gigabyte, the cheapest way to cut the bill is to stop logging full prompts and completions — which strips out precisely the context required to debug a failure. Pricing should not push you toward worse observability.
- 03
Phoenix to AX is a repricing event, not a tier upgrade
Phoenix self-hosted has no span or retention caps. The step to AX is a change of commercial model rather than a bigger version of the same plan, and teams report discovering that late.
Side by side8 ROWS
The matrix, wins and losses both
| Dimension | Arize AX | SouthBase |
|---|---|---|
| Pricing axes | Spans + GB, metered separately | One usage unit |
| Incentive to log full payloads | GB metering makes full logging more expensive | Payload size does not change the unit price |
| Seat cost | Unlimited users | Unlimited seats |
| ML drift / embeddings | Deepest in the category | Not a target — we do not build this |
| Claude token accounting | Generic LLM span attributes | Four Claude-native buckets, cache-hit ratio included |
| Self-host | Phoenix is self-hostable | Cloud only — collector runs in your infrastructure |
| Wire format | OpenInference — they author it | OpenTelemetry / OpenInference. The same instrumentation points at any OTLP backend. |
| Agent runs in your infrastructure | Phoenix can, AX cloud does not redact at the edge | Yes — Apache-2.0 collector, redaction before egress |
A marked cell is the stronger option on that row. We did not omit rows we lose.
Don't switch if…
When Arize AX is the right answer
If drift detection and embedding analysis are why you bought an observability tool, or you need an enterprise compliance posture today, Arize is the stronger choice and we will say so.
And if you do want to try ours
The integration is OpenTelemetry, so trying SouthBase does not mean removing Arize AX — you can point the same instrumentation at both and compare the output on your own traffic.
Other comparisons