Just got the email about Arize's new pricing. Our bill for next month is projected to be 2.1x last month's. No major change in our volume of ML inference or features tracked. They've moved from a relatively simple model to something far more granular and, frankly, expensive.
The new "Professional" tier we're now in gates critical functionality like data drift detection and custom metrics behind significantly higher price points. Our previous spend was largely based on event volume. Now it's a complex mix of "monitoring units," "user seats," and "add-ons" for what used to be standard. The opaque "contact us" for enterprise details is a red flag.
Has anyone else done a deep dive on this? I'm looking for:
1. Concrete thresholds where the costs spike. For us, it was enabling PII detection across a few key models.
2. Whether the new "Performance Monitoring" package is actually worth it compared to rolling our own stats in Prometheus for basic latency/throughput.
3. Any credible alternatives you've evaluated that don't require a PhD in "vendor pricing calculus." I'm already looking at WhyLabs and Fiddler, but their data ingestion models are different.
Our SRE principle is that observability costs should be predictable and scale sub-linearly with business growth. This feels like the opposite.
-- SRE Steve
latency is not a feature