# Culpa vs Traceloop

> Traceloop meters spans and answers whether output was good. Culpa prices the call and answers whether the customer was profitable. Both, not either.

URL: https://getculpa.com/culpa-vs-traceloop
Last reviewed: 2026-08-03

## Answer
Traceloop is an LLM reliability platform built on OpenTelemetry, and its own pricing meters spans rather than spend. Culpa, a local-first LLM cost, margin, and forecast ledger, prices every call from a versioned price book and forecasts next month. Traceloop answers whether the output was good. Culpa answers what it cost and whether that customer pays enough.

## Why this happens
These two barely overlap, and saying so is more useful than manufacturing a fight. Traceloop is about quality: faithfulness, relevance and safety checks on real traffic, drift detection, and custom evaluators trained on your own examples. Nothing in its published feature set claims cost tracking, and its pricing page meters SPANS rather than dollars. Its FAQ defines a span as every call to an LLM and every call to a vector database, which is the detail worth pausing on. A retrieval request spends one model call and several vector queries, so a Traceloop meter climbs with retrieval fan-out while a provider bill climbs with tokens. Those are two curves and neither predicts the other. Traceloop was acquired by ServiceNow in March 2026 and says it will continue to sell the Traceloop functionality, so this is a comparison rather than a migration. The reason to run both is that a good answer which loses money is still a problem, and neither tool sees the other half.

## What this usually looks like
- You know your evaluation scores and you can't name what a passing answer costs to produce.
- Your observability bill grew when you added retrieval, and your model bill didn't move.
- A quality regression and a cost regression look identical on your dashboards, which is to say invisible.
- You can say which prompt version scores best and not which one earns most.
- Retention on your free tier is measured in hours and nobody chose that number.

## Common mistakes
- Treating an evals platform as a cost tool because it sees every call. Why it hurts: Seeing a call and pricing it are different jobs. Traceloop meters spans and publishes no model rate card. Do instead: Keep the evaluator for quality and price the same traffic against a versioned price book.
- Sizing a span budget from your token volume. Why it hurts: Vector queries are spans too, so a retrieval app burns spans far faster than it burns tokens. Do instead: Count spans per request in your own traces before picking a tier.
- Reading the ServiceNow acquisition as a reason to leave. Why it hurts: Traceloop says it will honour customer obligations and keep selling the functionality, and OpenLLMetry stays open source. Do instead: Ask about the roadmap, the same as you would of any vendor that changed hands.
- Letting a free tier decide how far back you can look. Why it hurts: 24 hours of retention answers what broke this morning and nothing at all about this quarter. Do instead: Decide the lookback your cost work needs, then choose the tier that matches it.

## Self-check
- Count the spans one typical request produces, model calls and vector queries together.
- Multiply that by monthly requests and compare it to the tier you're on.
- Ask your evals tool what your best-scoring prompt version costs per call.
- Check your retention in hours, then ask which cost question needs more than that.
- Whatever goes unanswered is the gap, and the gap is the whole comparison.

## What fits inside the free tier, in spans and in dollars
Traceloop's free tier publishes 50,000 spans a month, and its FAQ counts every model call and every vector database call as a span. Take a retrieval request that makes 1 model call and 5 vector queries, so 6 spans, and price the same traffic on Claude Haiku 4.5 at $1.00 and $5.00 per million from the price book, effective 2026-07-02. Request shape is modelled. The rates and the allowance are published.
  50,000 spans / 6 spans per request = 8,333 requests a month inside the free tier
  2,000 input tokens x $1.00 per million = $0.002 per request
  500 output tokens x $5.00 per million = $0.0025 per request
  8,333 requests x $0.0045 = $37.50 of model spend a month
$37.50 of provider spend exhausts a 50,000-span allowance. Drop the retrieval step and the same $37.50 uses a sixth of it. The meter tracks your architecture rather than your bill, which is exactly why a span count can't answer a spend question.

## Cost figures
Every figure carries its confidence and its source. For each provider-reported figure below, its own source line names the published record it came from, the date that record was read, and any aggregation applied to it.
- $37.50 per month, model spend of the traffic that exhausts Traceloop's published 50,000-span free tier, on a retrieval shape [calculated] Source: Span allowance and span definition from traceloop.com/pricing, read 2026-08-03. 6 spans per request (1 model call plus 5 vector queries) gives 50,000 / 6 = 8,333 requests. At 2,000 input and 500 output tokens on Claude Haiku 4.5, real rates of $1.00 and $5.00 per million from the price book effective 2026-07-02: 8,333 x ($0.002 + $0.0025) = $37.50. Request shape and token counts are modelled. The rates and the allowance are published.
- 24 hours, published data retention on Traceloop's Free Forever tier [provider-reported] Source: traceloop.com/pricing, read 2026-08-03. The Free Forever card lists $0 a month, up to 50K spans a month, and 24 Hours Data Retention. Enterprise lists Custom Data Retention. Reported because retention is the ceiling on every backward-looking cost question, not because 24 hours is wrong for a free tier.

## FAQ
Q: Does Traceloop track LLM cost?
A: Its published feature set doesn't claim to, as read on 2026-08-03. Traceloop sells monitoring, evaluation, drift detection and prompt management, and its pricing meters spans rather than dollars. That's a statement about what it publishes rather than about what any dashboard might display.

Q: What counts as a span, and why does that matter for cost?
A: Traceloop's own FAQ says every call to an LLM and every call to a vector database is a span. It matters because a span count then tracks retrieval fan-out rather than token spend. Adding a reranker raises your observability bill and leaves your provider bill untouched, and the reverse is equally true.

Q: Is Traceloop shutting down after the ServiceNow acquisition?
A: No. Its March 2026 post says it intends to honour all obligations to customers and continue to sell the Traceloop functionality, and that OpenLLMetry stays open source. Its pricing page and free tier were both live on 2026-08-03. Ask about the roadmap the way you would of any vendor that changed hands.

Q: Do I have to choose between Traceloop and Culpa?
A: No, and this is one of the pairs where choosing would be the mistake. Traceloop answers whether the output was good. Culpa answers what producing it cost and whether that customer is profitable. A high-scoring answer that loses money is still a problem, and neither tool sees the other half.

Q: How long does Traceloop keep my data?
A: 24 hours on the published free tier and Custom on Enterprise, from its pricing page read 2026-08-03. Note a contradiction on that same page worth checking against your own account: the free tier card says up to 5 seats while the FAQ below it says there's no seat limit.

## Sources
- Traceloop: https://www.traceloop.com
- Traceloop pricing: https://www.traceloop.com/pricing
- Traceloop joining ServiceNow: https://www.traceloop.com/blog/traceloop-is-joining-servicenow

Run the free Cost Leak Scan: https://app.getculpa.com/scan?source=pseo&slug=culpa-vs-traceloop&cluster=competitor

Machine-readable index of every guide: https://getculpa.com/api/pages
Human-readable index of every guide: https://getculpa.com/guides
Site overview: https://app.getculpa.com/llms.txt

Privacy: 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.