What AI testing costs and why vendors rarely publish prices
By AI Testing Vendors · updated 2026-11-09
Of the published entries in this directory, few print any price on their own website, so cost comes from quotes. Quotes are comparable only when every vendor receives the same scope document.
How many vendors publish a price
7 of the published entries print a price with a currency on their own website. The directory records only whether a price is published, never the amount.
| Company type | Entries | Price published on own site |
|---|---|---|
| Pure-play QA company | 28 | 1 |
| IT services firm with a QA practice | 18 | 2 |
| AI assurance and audit specialist | 4 | 0 |
| Security firm with AI testing | 20 | 3 |
| Crowdtesting platform | 3 | 1 |
| Data and evaluation vendor | 12 | 0 |
Why prices are rarely published
AI testing work is scoped by the product type, the number of model versions under test, the size of the evaluation set and whether human reviewers grade the output. Each of these changes the effort, so vendors quote per engagement.
What drives the quote
- The engagement model: a one-time audit, a fixed-scope project, a dedicated team or a retainer.
- The size and ownership of the golden set: building one costs more than running an existing one.
- Human review: rubric grading by people costs more per item than automated scoring.
- Adversarial depth: manual red teaming takes longer than running a scanner.
- Regulatory output: evidence packs for a conformity assessment add documentation time.
Engagement models vendors name
| Engagement model | Entries naming it on the AI page |
|---|---|
| Staff augmentation | 3 |
| Dedicated team | 3 |
| Managed testing | 9 |
| Fixed-scope project | 8 |
| One-time audit or assessment | 25 |
| Retainer or subscription | 3 |
| None named | 51 |
Scope document to send
- 01
Product
Product type, model provider, release date and the user-facing channels.
- 02
Risks
The failure modes you care about: wrong answers, harmful output, data leakage, bias, latency.
- 03
Volume
Number of prompts or cases per run, number of runs, and whether a golden set exists.
Send the same document to every vendor so the quotes cover the same work.
Related pages
- LLM output evaluation
- RAG pipeline validation
- Prompt and model regression testing
- Red teaming and adversarial testing
- Pure-play QA company
- IT services firm with a QA practice
- AI assurance and audit specialist
- Security firm with AI testing
- All companies
- How entries are verified
- How to choose a vendor to test an AI product
Questions
What drives the cost of testing an AI product?
The product type, the size and ownership of the evaluation set, the share of human grading, the depth of adversarial testing, the number of model or prompt versions, and any regulatory documentation. The engagement model also changes how cost is billed.
Why does testing an AI application usually cost more than standard software testing?
Outputs vary between runs, so each case is run several times and scored instead of checked once. Building evaluation sets, grading answers by people and adversarial testing add work that ordinary functional testing does not need.
Why do AI testing companies rarely publish their prices?
Effort depends on scope details that differ for every product, so vendors quote per engagement. 7 of 88 published entries print a price on their own site.
How do you scope an AI testing engagement?
Describe the product and model, the risks you care about, the volume of cases and runs, whether an evaluation set exists, and what the report must contain. Agree acceptance thresholds before testing starts.