Guides / culpa vs arize
Culpa vs Arize
Arize AX is an agent observability and evaluation platform metered on trace spans and ingestion volume, with Phoenix as its open-source counterpart. Culpa, a local-first LLM cost, margin, and forecast ledger, prices each provider call from a versioned price book, attributes it to a customer against revenue, and forecasts next month with a range.
Why this happens
Arize meters two things at once and neither is money. AX Free covers 25k trace spans a month and 1 GB of ingestion with 15-day retention. AX Pro is $50 a month for 50k spans and 10 GB with 30-day retention. Both are SaaS, both give unlimited users and unlimited evals, and Phoenix is the open-source option if you'd rather run it yourself. The shape to notice is that a span count and an ingestion volume can both be growing while your model spend falls, because richer traces and smaller prompts pull in opposite directions. Arize is built for the question of whether an agent behaved, which is a real and hard question. It doesn't publish margin against revenue or a forecast of spend, and 15 days of retention on the free tier is the practical limit on how far back any cost question can reach.
What this usually looks like
- Your span count doubled after a refactor and your model bill didn't move.
- You can prove an agent behaved and not what the run cost.
- Retention expired before the monthly cost review asked its question.
- Ingestion volume grew because traces got richer, not because usage did.
- Nobody can name which customer's agent runs are the expensive ones.
Free, no card, no account
Run the free Cost Leak Scan
It shows your most expensive conversation before you install anything.
Mistakes that cost the most
| Mistake | Why it hurts | Do instead |
|---|---|---|
| Using span count as a stand-in for spend. | Spans track how you instrumented the code. Spend tracks tokens. Those move independently. | Price the calls from a rate book and keep spans for behaviour. |
| Picking a tier on span count alone. | Arize meters ingestion volume too, so a verbose trace can hit the GB ceiling long before the span ceiling. | Check both meters against your own traffic before committing. |
| Assuming an open-source counterpart means the same product. | Phoenix and AX are published as different things, and the retention and hosting differ. | Decide which one you're actually running before you build reporting on it. |
Run this check tonight
- Count the spans one agent run emits, then check it against your tier.
- Check your ingestion volume separately from your span count.
- Ask what a single agent run costs in provider spend, not in spans.
- Find your retention window and compare it to your reporting cycle.
When instrumentation moves the meter and spend stays still
Arize AX Free publishes 25k trace spans a month. Take an agent that made 5 spans per run before someone instrumented the retrieval steps, and 12 after, with identical model calls throughout. Rates and allowances are published. The span shape is modelled.
The meter measured the instrumentation, not the workload. That's fine for observability billing and it's why a span-based tool can't answer a spend question, which needs tokens priced against a rate book rather than events counted.
Every number, with its confidence and source
| Figure | What it means | Confidence | Source |
|---|---|---|---|
| $50 per month | Arize AX Pro's published price | provider-reported | arize.com/pricing, read 2026-08-03. AX Pro covers 50k trace spans a month, 10 GB ingestion volume and 30-day retention on SaaS with unlimited users. AX Free covers 25k spans, 1 GB and 15-day retention. Phoenix is published separately as open source. |
What a generic answer can’t know
Arize is aimed squarely at agent behaviour, and it publishes Phoenix under an open-source licence for teams who want to self-host, so privacy and control aren't the argument here. The argument is scope. Arize tells you what an agent did, how well it did it, and how much telemetry that generated. Culpa tells you what the underlying calls cost in exact decimal from a versioned price book, what the customer who triggered them pays you, and what next month looks like with a range. Neither of those last two appears anywhere in Arize's published feature set, and that's the whole comparison. Keep the one that watches behaviour and add the one that watches the money.
Questions founders ask next
What does Arize AX cost?
AX Free covers 25k trace spans a month, 1 GB of ingestion and 15-day retention, with unlimited users and evals. AX Pro is $50 a month for 50k spans, 10 GB and 30-day retention. Both are SaaS. Published on arize.com/pricing, read 2026-08-03.
Does Arize track LLM cost?
Its published feature set is built around traces, spans and evaluation rather than a rate book. Nothing on its pricing or product pages claims margin against revenue or a forecast of spend, as read on 2026-08-03. Treat any cost column you see as attribution rather than as a priced ledger.
What's the difference between Arize AX and Phoenix?
Arize publishes AX as its managed AI engineering platform and Phoenix as open-source observability and evals. They're presented as separate products with different hosting, so decide which one you're running before building reporting on either.
Can I run Arize and Culpa together?
That's the intended shape. Arize answers whether the agent behaved. Culpa prices the calls it made, sets them against what the customer pays, and forecasts the month. The two read the same traffic and answer different questions.
On your infrastructure
Culpa runs on your infrastructure. Your prompts and responses never leave it. Culpa counts calls to run your plan, and it fails open, so if it ever breaks your app keeps running.
How Culpa works
Find the culprit. Not just the total.
Your dashboard shows what you spent. It stops short of who spent it. Culpa shows the conversation, the user and the feature behind it.
Your prompts stay local.
Culpa runs on your own infrastructure. What you send to a model reaches us at no point.
Every dollar has a name.
Follow any charge to the conversation, the user, the feature and the customer behind it.
See the bill before it lands.
Cost your next feature before you ship it. You get the likely bill and the worst case, at best, median, p90 and p99.
Three steps to your first answer.
Change one base URL.
Or drop in the Python or TypeScript library.
Find your most expensive conversation.
In the first session, not the first week.
Cost your next feature before you ship it.
Why the bill went up
Example dashboardCalls traced
418,209
across 3 projects
Spend this week
$378.41
+ $182 vs last week
Failed calls
312
74% retried, and you paid for all of them
+ $182 this week traced to one culprit
Spend over 14 days
Most expensive users
Next week forecast
Graded against reality. Accuracy shown as results land.
Free, no card, no account
Run the free Cost Leak Scan
It shows your most expensive conversation before you install anything.
Keep reading
Sources: Arize, Arize AX pricing. Last reviewed 2026-08-03. Plain text version.