Does a company's founding year indicate how much experience it has specifically in AI system testing?

Founding year measures how old a company is. The service on this page is recorded as evaluation of machine learning and language model systems, and founding years in this group run from 1997 through 2020, so most of these companies were operating long before the systems this service examines. No profile records when AI system testing joined a company's list, which leaves a long practice and a recent addition reading the same way here.

If a company's profile lists the industries it has worked in, does that show it can judge whether an AI model's output is actually correct?

No. An industry listed on a profile records where a company has delivered projects. It does not say whether anyone on the team can tell a correct model answer from an incorrect one, which takes domain knowledge a delivered-projects list does not capture. A functional defect shows itself: a button that fails to respond. A wrong answer from a model is visible only to someone who already knows the correct answer in that domain, and this catalog has no field for whether a company employs people with that judgment.

Does a published hourly rate tell a buyer what evaluating an AI model would cost?

Not on its own. An hourly rate describes time spent, while evaluating a model is commonly scoped as a run across a dataset, repeated each time the model changes, a unit an hourly figure does not capture. Sixteen of the 22 companies in this group publish an hourly band, and each one is blended rather than split by task. The catalog records no per-run or per-dataset price for any company here, so a published rate does not describe what a model evaluation would cost.

Best AI system testing companies in 2026

QA Rating is published independently, and its publisher owns no company listed in this catalog.

22 testing companies in this catalog list AI system testing among their services. They are grouped below by what each is best suited for, in 9 groups. 2 of them hold a certificate confirmed by a register entry and 16 publish a rate card. Composition and order last changed on September 6, 2026.

22 companies. Ordered by the number of verified reviews on external platforms, highest first. Companies with an equal count are ordered by name.

The companies

AI system testing: 22 companies, grouped by what they are best suited for.
#	Company	Frameworks	Case studies	Published rates	Reviews
Automation first Test automation is offered as a service, and at least one published case study reports a coverage or regression time metric.
1	QA Madness	Cypress, Playwright, Selenium, Appium, Robot Framework, WebdriverIO, k6, Apache JMeter	2	Blended $25-49/hclutch.co	44
2	QASource	Salesforce, Selenium, Playwright, Cypress, Apache JMeter, Appium, Espresso, XCUITest	1	Blended $25-49/hclutch.co	29
3	BugRaptors	Selenium, WebdriverIO, Cypress, Playwright	2	Blended up to $25/hclutch.co	26
4	Luxe Quality	Salesforce, Selenium, Cypress, Playwright, Cucumber, WebdriverIO, Robot Framework, TestComplete, Appium, Espresso, XCUITest, Postman, SoapUI, REST Assured	2	Blended $25-49/hclutch.co	21
5	QAwerk	Selenium, Robot Framework, Appium, TestNG, JUnit	3	Blended $25-49/hclutch.co	13
6	QAble	Selenium, Cypress, Playwright, Appium, WebdriverIO, k6, Apache JMeter, Gatling, Postman, REST Assured	4	Blended up to $25/hclutch.co	8
7	Frugal Testing	Microsoft Dynamics 365, ServiceNow, Selenium, TestNG, Apache JMeter	1	Blended up to $25/hclutch.co	7
8	Moolya	Selenium	4	Blended $25-49/hclutch.co	7
9	Codoid	Selenium	2	Blended up to $25/hclutch.co	5
10	2i	Oracle applications, Microsoft Dynamics 365, Salesforce, SAP, ServiceNow, Workday	3	Not published	0
11	ImpactQA	Oracle applications, SAP, Tricentis Tosca	5	Not published	0
12	QualityAI	Not published	2	Not published	0
Embedded team model The dedicated team engagement model is offered, and at least one published case study.
13	Testlio	Not published	5	Blended $25-49/hgoodfirms.co	76
1	QA Madness	Cypress, Playwright, Selenium, Appium, Robot Framework, WebdriverIO, k6, Apache JMeter	2	Blended $25-49/hclutch.co	44
14	TestFort	SAP, Selenium, Appium	5	Blended $25-49/hclutch.co	32
2	QASource	Salesforce, Selenium, Playwright, Cypress, Apache JMeter, Appium, Espresso, XCUITest	1	Blended $25-49/hclutch.co	29
5	QAwerk	Selenium, Robot Framework, Appium, TestNG, JUnit	3	Blended $25-49/hclutch.co	13
6	QAble	Selenium, Cypress, Playwright, Appium, WebdriverIO, k6, Apache JMeter, Gatling, Postman, REST Assured	4	Blended up to $25/hclutch.co	8
10	2i	Oracle applications, Microsoft Dynamics 365, Salesforce, SAP, ServiceNow, Workday	3	Not published	0
11	ImpactQA	Oracle applications, SAP, Tricentis Tosca	5	Not published	0
12	QualityAI	Not published	2	Not published	0
15	QualityMinds	Not published	1	Not published	0
Nearshore for the EU At least one delivery location in the EU or the EEA, or in Ukraine, Georgia, Serbia or Moldova.
13	Testlio	Not published	5	Blended $25-49/hgoodfirms.co	76
16	Global App Testing	Not published	4	Not published	66
1	QA Madness	Cypress, Playwright, Selenium, Appium, Robot Framework, WebdriverIO, k6, Apache JMeter	2	Blended $25-49/hclutch.co	44
14	TestFort	SAP, Selenium, Appium	5	Blended $25-49/hclutch.co	32
4	Luxe Quality	Salesforce, Selenium, Cypress, Playwright, Cucumber, WebdriverIO, Robot Framework, TestComplete, Appium, Espresso, XCUITest, Postman, SoapUI, REST Assured	2	Blended $25-49/hclutch.co	21
5	QAwerk	Selenium, Robot Framework, Appium, TestNG, JUnit	3	Blended $25-49/hclutch.co	13
17	AskYourQA	Selenium, Playwright, Appium, Espresso, Cypress	1	Blended $25-49/hclutch.co	7
9	Codoid	Selenium	2	Blended up to $25/hclutch.co	5
15	QualityMinds	Not published	1	Not published	0
18	Qytera Quality GmbH	Playwright, Selenium, Robot Framework, Apache JMeter, Postman, Cypress	7	Not published	0
Performance engineering Performance testing is offered as a service, and at least one published case study reports a throughput or latency metric.
14	TestFort	SAP, Selenium, Appium	5	Blended $25-49/hclutch.co	32
2	QASource	Salesforce, Selenium, Playwright, Cypress, Apache JMeter, Appium, Espresso, XCUITest	1	Blended $25-49/hclutch.co	29
3	BugRaptors	Selenium, WebdriverIO, Cypress, Playwright	2	Blended up to $25/hclutch.co	26
18	Qytera Quality GmbH	Playwright, Selenium, Robot Framework, Apache JMeter, Postman, Cypress	7	Not published	0
Security focus Security testing is offered, together with either penetration testing or a verified ISO/IEC 27001 certificate, and at least one published case study.
1	QA Madness	Cypress, Playwright, Selenium, Appium, Robot Framework, WebdriverIO, k6, Apache JMeter	2	Blended $25-49/hclutch.co	44
5	QAwerk	Selenium, Robot Framework, Appium, TestNG, JUnit	3	Blended $25-49/hclutch.co	13
Accessibility specialists Accessibility testing is offered as a service, and at least one published case study reports an accessibility conformance metric.
19	Vervali Systems	Not published	1	Blended $25-49/hclutch.co	11
Crowdtesting at scale The crowdtesting engagement model is offered, and a published tester pool of 10,000 people or more, with a source.
16	Global App Testing	Not published	4	Not published	66
Device lab coverage A published device lab of 100 real devices or more, with a source.
20	PLUS QA	pytest, TestNG, JUnit, Selenium, Appium, Postman	Not published	Blended $50-99/hclutch.co	8
Nearshore for the USA At least one delivery location in Canada, Mexico, Central America or South America.
2	QASource	Salesforce, Selenium, Playwright, Cypress, Apache JMeter, Appium, Espresso, XCUITest	1	Blended $25-49/hclutch.co	29
Other companies in this axis No applicability tag met its criterion for these profiles.
21	Kualitatem	Not published	Not published	Blended $25-49/hclutch.co	9
22	Testrig Technologies	Salesforce, Selenium, Cypress, Playwright, Tricentis Tosca, Katalon, Appium, TestComplete, Robot Framework, REST Assured, Postman, Apache JMeter, SoapUI, k6, Cucumber	Not published	Blended up to $25/hclutch.co	7

What "AI testing" mixes together

A company appears in this group of 22 when its profile records AI system testing as a service, defined in this catalog as evaluation of machine learning and language model systems: output quality, regression suites and guardrail checks. That service claim comes from a vendor's own site, and the same three words cover two different things in the market. One is testing where the system under test is itself an AI or ML product, such as a model or an agent. The other is running an ordinary testing process with AI-powered tools on software that has nothing to do with AI. A service field built from a vendor's own wording cannot tell the two apart on its own.

Case study titles are one place where the difference becomes partly visible. In this group of 22, the word "AI" appears in a published case study title for 3 companies: QAble, with an AI game platform case and an AI platform case; TestFort, with a case about an AI-powered wearable; and QualityMinds, with a case about an AI workshop for a named client. Another 16 companies in this group have at least one published case study with no "AI" in any title, describing instead the client, the industry or the testing activity. The remaining 3 companies in this group have no published case study recorded at all. A title naming "AI" records a different fact than a metric on an AI system's own output: none of the case study data collected for this page reports whether the figures behind those three titles measure a model's output specifically or the application around it.

Where the group splits further

Nine applicability tags cut this group of 22 into overlapping subsets. Automation first is the largest, 12 of the 22: QA Madness, QASource, BugRaptors, Luxe Quality, QAwerk, QAble, Frugal Testing, Moolya, Codoid, 2i, ImpactQA and QualityAI. Nearshore for the EU and embedded team model follow, tied at 10 of the 22 each. Nearshore for the EU covers Testlio, Global App Testing, QA Madness, TestFort, Luxe Quality, QAwerk, AskYourQA, Codoid, QualityMinds and Qytera Quality GmbH. Embedded team model covers Testlio, QA Madness, TestFort, QASource, QAwerk, QAble, 2i, ImpactQA, QualityAI and QualityMinds. Performance engineering covers 4 of the 22: TestFort, QASource, BugRaptors and Qytera Quality GmbH.

Security focus has 2 members in this group of 22, QA Madness and QAwerk, and both also carry automation first, nearshore for the EU and embedded team model at the same time, four tags in total. Checked against the tag count recorded for every one of the 22 companies, QASource also reaches four tags, carrying embedded team model, automation first, performance engineering and nearshore for the USA instead of security focus; these three companies share the highest count on this page. The next-highest count, three tags, belongs to TestFort alone: embedded team model, performance engineering and nearshore for the EU. The four remaining tags each have one member: accessibility specialists belongs to Vervali Systems, crowdtesting at scale belongs to Global App Testing, device lab coverage belongs to PLUS QA, and nearshore for the USA belongs to QASource. Two companies in this group of 22, Kualitatem and Testrig Technologies, carry none of the nine tags: checked against every tag's criteria, explained on the methodology page, neither profile matches any of the nine.

What the profiles do not show

A verified certificate is recorded for 2 of the 22 companies in this group, and the certificate field carries no verified entry for the other 20. Neither of the two holders reaches this composition through that field: none of the nine tags on this page counts a certificate on its own. The count records what this catalog could confirm from a public register; a company may hold an audit that no such register carries.

Whether a company signs a BAA or a GDPR data processing agreement is unrecorded for all 22 companies in this group. AI testing work often involves production data or model prompts passing through a vendor's own environment before a defect or a guardrail failure is logged, and no profile in this group has a published answer on that point. None of the fields collected for this page describe how a company evaluates a model's output when that model changes its own behaviour between releases, or how it handles a guardrail failure once one is found. That is what a call with a shortlisted company still has to cover.

How this ranking was put together

Membership is computed from the profile, not chosen: companies appear here when they list AI system testing among their services, as their own profile states. Position in a ranking cannot be bought. Paid options are limited to the extended profile, labelled sponsor slots and lead forms, and none of them affect tags or order.

The table compares the fields that decide this axis: frameworks, case studies, published rates, plus the number of verified reviews on external platforms. Rates, minimum project size and onboarding time carry the source they came from next to the value.

Automation first: Test automation is offered as a service, and at least one published case study reports a coverage or regression time metric.
Embedded team model: The dedicated team engagement model is offered, and at least one published case study.
Nearshore for the EU: At least one delivery location in the EU or the EEA, or in Ukraine, Georgia, Serbia or Moldova.
Performance engineering: Performance testing is offered as a service, and at least one published case study reports a throughput or latency metric.
Security focus: Security testing is offered, together with either penetration testing or a verified ISO/IEC 27001 certificate, and at least one published case study.
Accessibility specialists: Accessibility testing is offered as a service, and at least one published case study reports an accessibility conformance metric.
Crowdtesting at scale: The crowdtesting engagement model is offered, and a published tester pool of 10,000 people or more, with a source.
Device lab coverage: A published device lab of 100 real devices or more, with a source.
Nearshore for the USA: At least one delivery location in Canada, Mexico, Central America or South America.

The criteria for this axis are on the methodology page, together with field definitions and the rule for editorial order. Changes of order are recorded in the changelog.

Questions

Does a company's founding year indicate how much experience it has specifically in AI system testing?: Founding year measures how old a company is. The service on this page is recorded as evaluation of machine learning and language model systems, and founding years in this group run from 1997 through 2020, so most of these companies were operating long before the systems this service examines. No profile records when AI system testing joined a company's list, which leaves a long practice and a recent addition reading the same way here.
If a company's profile lists the industries it has worked in, does that show it can judge whether an AI model's output is actually correct?: No. An industry listed on a profile records where a company has delivered projects. It does not say whether anyone on the team can tell a correct model answer from an incorrect one, which takes domain knowledge a delivered-projects list does not capture. A functional defect shows itself: a button that fails to respond. A wrong answer from a model is visible only to someone who already knows the correct answer in that domain, and this catalog has no field for whether a company employs people with that judgment.
Does a published hourly rate tell a buyer what evaluating an AI model would cost?: Not on its own. An hourly rate describes time spent, while evaluating a model is commonly scoped as a run across a dataset, repeated each time the model changes, a unit an hourly figure does not capture. Sixteen of the 22 companies in this group publish an hourly band, and each one is blended rather than split by task. The catalog records no per-run or per-dataset price for any company here, so a published rate does not describe what a model evaluation would cost.

Previous versions

Each year is frozen as it stood when the snapshot was taken, with the field values recorded at that date. This page always shows the current version.

The 2026 list: 18 companies, taken 2026-09-03. 4 entered and 0 left since.

Other axes these companies appear on

Industry: Ecommerce 18 of these 22 companies also stand there
Industry: Banking 16 of these 22 companies also stand there
Delivery location: Europe 16 of these 22 companies also stand there
Industry: Healthcare 15 of these 22 companies also stand there

How to choose a vendor on this axis

Published by QA RatingPublished on September 3, 2026Updated on September 5, 2026

Every fact about a company comes from a listed source. A field without a source stays empty. A certificate counts as confirmed only when a registry, certificate or auditor report backs it; a certificate the company only claims about itself is published separately, under that label. Vendors can request a correction at hello@qa-rating.com. A correction is applied when it comes with a public source.