Skip to content

Benchmark · 10 resumes · 7 October 2026

HireLayer vs Textkernel, Affinda and RChilli on ten real-world resume layouts

We wrote ten fictional resumes, so the right answer to every field is known, and ran them through four resume parsers. Scans, a phone photo, an infographic, Europass, Word tables: HireLayer is the only one that got both the work history and the language levels right.

  • Every field checked against a known answer
  • Raw API JSON, not on-screen previews
  • All ten test files free to download
  • Round 2 scoring rules fixed before any run

Scorecard

10 resumes · raw JSON

Language levels correct

out of 28
  • HireLayer26
  • Affinda12
  • Textkernel0
  • Textkernel LLM0

Positions fully correct

out of 41
  • HireLayer38
  • Affinda30
  • Textkernel27
  • Textkernel LLM38

RChilli ran on the first five resumes only: see every score.

Test resumes
107 layouts, French and English
Parsers
4Textkernel run twice: default and LLM
Positions to find
41Employer, title and dates
Languages to level
28CEFR, scores and plain words

The scores, field by field

A position counts only when the employer, the job title and both dates match the resume. A language counts when the level matches what is written, or its plain reading for words such as “fluent” or “mother tongue”.

Benchmark scores for HireLayer, Affinda, Textkernel (default and LLM Parser) and RChilli
What we checkedHireLayer Affinda logo Textkernel logoDefault settingsTextkernel logoLLM Parser onRChilli logoRound 1 only
Language levels correct28 languages across 10 resumes, from CEFR codes to “mother tongue”, IELTS, JLPT and Europass grids26/2812/287 wrong levels0/28No level field0/28No level field0/13No level field
Positions fully correctEmployer, job title, start and end dates all match the resume38/4130/4127/4138/4110/18
Resumes with no correct positionThe whole work history came back missing, undated or garbled02Infographic, phone photo2Infographic, Europass02 of 5Scan, infographic
Name, email or phone wrong or missingResumes where the candidate cannot be identified or reached from the parsed profile12522 of 5
French contract types typedApprenticeship, internship, temp work and fixed-term contract in round 14/42/4Apprenticeship and temp work filed as internships0/4UNSPECIFIED0/4UNSPECIFIEDNo field
ROME occupation codesFrance Travail job classification, with a scoreYesNoNoNoNo
Gender returned though never writtenNone of the ten resumes states a genderNeverNeverYes4 of 5 in round 1YesOn all 10 resumesNever

Totals cover the ten resumes. RChilli ran on the first five only, so its scores are out of 18 positions and 13 languages. Textkernel offers an optional LLM Parser in its demo; we report both runs. Green marks the best result on a row, red a clear miss.

Recruiters filter on “C1”, not on “English”

A parser that returns the language without the level leaves the hardest part to a human. HireLayer returned a level for all 28 languages, mother tongues included, and got 26 right. Textkernel and RChilli have no level field at all.

On the resume · CV5, right-hand column
Italian Native · English Fluent (C1) · French Upper-intermediate (B2)
  • Scores and grids: HireLayer converted an IELTS 7.5, a DELF B1 and a Europass self-assessment grid into CEFR levels.
  • Plain words: on CV2, “English (bilingual)” and “Yoruba (native)” both became C2, where Affinda dropped Yoruba altogether.
  • Affinda: 12 right and 7 wrong, such as B2 English for a candidate who wrote “English (IELTS 7.5)”, a C1 score.

The scales behind those labels are in our guide to language levels on a resume.

3 languages, 3 right levels

HireLayer · raw JSONCV5
"languages": [
  { "language": "Italian", "level": "Native or Bilingual (C2)" },
  { "language": "English", "level": "Full Professional Proficiency (C1)" },
  { "language": "French", "level": "Professional Working Proficiency (B2)" }
]

Scans, photos and designer CVs are where parsers split apart

On the clean two-page executive CV, HireLayer, Affinda and Textkernel all found the nine positions. These four layouts are common in real application flows, and they decide whether a candidate profile is usable or has to be retyped.

  • First page of CV6, Karim Haddad

    CV6 · Karim Haddad

    A phone photo of a printed CV

    Shot at an angle, with a shadow across the page.

    • Right

      HireLayer · Contact, 3/3 positions, 2/2 diplomas, two machinery licences and an electrical clearance as three separate certifications

    • Wrong

      Affinda · Phone missing, the first word of the job title dropped, employers cut in half

    • Partly right

      Textkernel · Email cut to “karim.haddad@example” by default, one of the two diplomas missing in both modes

  • First page of CV4, Thomas Lefèvre

    CV4 · Thomas Lefèvre

    A scanned page

    Image only, slightly rotated, with scanner noise.

    • Right

      HireLayer · Email, accents, 3/3 positions with their cities, B2 and A2 levels

    • Partly right

      Affinda · Positions right, but B2 English read as C1 and basic German as B1

    • Wrong

      Textkernel · Email read as “thornasiefevre@example.org” in both modes

    • Wrong

      RChilli · No position and no diploma returned

  • First page of CV5, Sofia Marchetti

    CV5 · Sofia Marchetti

    A designer infographic

    Dates in a timeline column, written “03.2021 — now”.

    • Right

      HireLayer · Current role, dates, both schools and over 10 years of experience

    • Wrong

      Affinda · Every position ends the month it starts: 14 months of experience in total

    • Wrong

      Textkernel · No dates by default, so “an entry-level candidate”; right with the LLM Parser

    • Wrong

      RChilli · No position, no school, last name lost

  • First page of CV10, Lucie Fontaine

    CV10 · Lucie Fontaine

    A Europass CV

    The EU template, with a self-assessment grid for languages.

    • Right

      HireLayer · 3/3 positions, 2/2 diplomas, a level for all 3 languages read from the grid

    • Partly right

      Affinda · Positions right; the language grid’s headings returned as skills

    • Wrong

      Textkernel · By default: no name, no phone, no position, no diploma

The infographic, as each API returned it

Dates in a timeline column are a classic trap: lose them and the candidate looks like a junior. Sofia has been a designer since 2015, leads design at BlaBlaCar and studied in Milan and Venice.

On the resume · CV5, timeline
03.2021 — now · Lead Product Designer · BlaBlaCar · Paris
01.2018 — 02.2021 · Freelance UX/UI Designer
MSc Digital and Interaction Design · Politecnico di Milano · 2013 — 2015

Current role, dates and both schools

HireLayer · raw JSONCV5
"info_candidate": {
  "full_name": "SOFIA MARCHETTI",
  "job_title": "Lead Product Designer",
  "experience_level": "More than 10 years"
},
"work_experiences": [
  {
    "company_name": "BlaBlaCar",
    "job_title": "Lead Product Designer",
    "start_date": "2021-03-01",
    "end_date": null,
    "currently_active": true,
    "work_experience_city": "Paris"
  },
  {
    "company_name": "Self-employed",
    "job_title": "Freelance UX/UI Designer",
    "contract_type": "Freelance",
    "start_date": "2018-01-01",
    "end_date": "2021-02-28",
    "currently_active": false
  }
  // trimmed
],
"educations": [
  { "degree_title": "MSc Digital and Interaction Design", "school_name": "Politecnico di Milano", "start_date": "2013-01-01", "end_date": "2015-12-31" },
  { "degree_title": "BA Graphic Design and Visual Communication", "school_name": "Università Iuav di Venezia", "start_date": "2010-01-01", "end_date": "2013-12-31" }
]

A profile you can't contact is a lost candidate

A misread email or a swapped name breaks deduplication and outreach. HireLayer returned the right name, email and phone on nine of the ten resumes. Textkernel, with default settings, got one of them wrong on five.

  • Scan (CV4)

    Textkernel read thomas.lefevre@example.org as “thornasiefevre@example.org”, in both modes. HireLayer, Affinda and RChilli read it right.

  • Phone photo (CV6)

    Affinda returned no phone number. Textkernel cut the email to “karim.haddad@example” with default settings.

  • Europass (CV10)

    Textkernel returned neither the name nor the phone number with default settings.

CV8 · written “NGUYEN Thi Lan”

Family name first

Expected: last name Nguyen, first name Lan (or “Thi Lan”).

ParserFirst nameLast nameResult
HireLayerThi LanNGUYENRight
AffindaNguyenLanWrong
TextkernelThiLanWrong
Textkernel, LLM ParserThiNguyenPartly right

Clean, normalized skills, without invented ones

Skills only help matching when they come back as clean values. On the backend engineer's resume, HireLayer returned 31 skills, all of them normalized with a domain and a subcategory. The others returned sentence fragments, missed core skills or added skills that are not on the resume.

Skills that were never on the resume

  • TextkernelStock Control “services controlling warehouse robots” on a backend engineer’s resume
  • TextkernelDialectical Behavior Therapy “dbt”, the data tool, on a data analyst’s resume
  • AffindaInstallation “interactive museum installations” on a product designer’s resume
  • AffindaFood Delivery “systems in logistics and food delivery” in a backend engineer’s summary
  • RChilliOnboarding Filed under HR specialists, from a designer’s client project “Qonto (onboarding for freelancers)”

Want the same normalization on your own skill lists? The Skills Normalizer maps free text to the same taxonomy.

31 of 31 skills normalized

HireLayer · raw JSONCV2 · backend engineer
"skills": [
  { "skill_title": "Go", "skill_type": "Software skill", "status": "normalized" },
  { "skill_title": "Kubernetes", "skill_type": "Software skill", "status": "normalized" },
  { "skill_title": "Apache Kafka", "skill_type": "Software skill", "status": "normalized" },
  { "skill_title": "Terraform", "skill_type": "Software skill", "status": "normalized" },
  { "skill_title": "gRPC", "skill_type": "Hard skill", "status": "normalized" },
  { "skill_title": "Spring Boot", "skill_type": "Software skill", "status": "normalized" }
  // domain and subcategory trimmed; 25 more skills trimmed
]

Apprenticeships, temp work, fixed-term contracts and ROME codes

An apprenticeship, an internship, temp work and a fixed-term contract don't weigh the same when you screen a French candidate. HireLayer typed all four, and it suggests ROME occupation codes with a confidence score. No other parser in the test returns them.

First ROME suggestion, with its score

  • M1718CV1 · 0.991

    For a digital marketing project manager

  • F1202CV6 · 0.985

    For a structural works site manager

  • H2510CV9 · 0.960

    For an industrial director

  • K1903CV10 · 0.951

    For an employment law specialist

No inferred gender

None of the resumes states a gender. Textkernel returned one anyway, a field you would have to strip before storing the profile in most hiring contexts. HireLayer, Affinda and RChilli did not.

Textkernel · raw JSONCV1
"PersonalAttributes": {
  "DateOfBirth": { "Date": "1995-03-14" },
  "DrivingLicense": "B",
  "Gender": "Female"
}

4 of 4 contract types

HireLayer · raw JSONCV1 and CV3
// CV1
{ "contract_type": "Apprenticeship", "start_date": "2017-09-01", "end_date": "2019-02-28" },
{ "contract_type": "Internship", "start_date": "2017-04-01", "end_date": "2017-07-31" },
// CV3
{ "contract_type": "Temporary assignment", "start_date": "2019-01-01", "end_date": "2022-12-31" },
{ "contract_type": "Fixed-term contract", "start_date": "2015-07-01", "end_date": "2015-08-31" }

How we kept the test checkable

HireLayer is our product, so this page is built to be verified rather than trusted. Both rounds ran on 7 October 2026, each parser through its own demo or sandbox, with default settings.

  • Known answers

    We wrote the resumes ourselves, for fictional people, so every expected value is set before the test.

  • Raw JSON only

    We scored the response each API returned, never the vendor's on-screen preview.

  • Rules written first

    For the second round, the expected values and scoring rules were written down before any parser saw the files, and all five resumes are reported.

ParserWhere we ran itSettings
HireLayerDashboard, Resumes workspace (same parser as POST /api/v4/resumes/parse)Default
AffindaResume Parser sandbox, EU regionDefault
TextkernelTx Platform demo, engine 9.22, API 10.13Default, then with “Enable LLM Parser”
RChilliResume Parser 8.0.0 demo, Frankfurt serverDefault, indexing off · round 1 only

One full HireLayer response

This is what your application receives for CV2, with long arrays trimmed where marked: contact details with a geocoded location, dated positions, EQF education levels, CEFR language levels, normalized skills and ROME codes.

Every field is described in the API documentation.

POST /api/v4/resumes/parse · 200 OKCV2 · Daniel Okafor
{
  "application_id": "app_1fa0ce2b-5",
  "request_id": "8bae6a12-bccc-4c6f-919b-d5bd75ec4695",
  "stored": true,
  "already_stored": false,
  "resume": {
    "info_resume": {
      "date_parsing": "2026-10-07T17:24:23",
      "language": "EN",
      "text": "…" // 2,645 characters, trimmed
      // url, face_url and face_url_expires_at trimmed
    },
    "info_candidate": {
      "full_name": "DANIEL OKAFOR",
      "last_name": "OKAFOR",
      "first_name": "DANIEL",
      "email": "daniel.okafor@example.com",
      "phone_number": null,
      "birth_date": null,
      "age": null,
      "availability_now": true,
      "availability_date": null,
      "driver_license": [],
      "job_title": "Senior Backend Engineer",
      "education_name": "MSc Advanced Computer Science",
      "education_level": "Level 7",
      "experience_level": "More than 10 years",
      "linkedin_url": "https://linkedin.com/in/danielokafor-dev",
      "github_url": "https://github.com/dokafor-dev",
      "other_urls": [],
      "location": {
        "country": "United Kingdom",
        "country_code": "GB",
        "region": "England",
        "department": null,
        "city": "London",
        "postal_code": null,
        "full_address": "London, United Kingdom",
        "latitude": 51.5073,
        "longitude": -0.127647
      },
      "mobility": { "can_work_in_other_cities": null, "other_cities": [] }
    },
    "educations": [
      {
        "degree_title": "MSc Advanced Computer Science",
        "school_name": "The University of Manchester",
        "description": null,
        "degree_type": "Level 7",
        "start_date": "2015-01-01",
        "end_date": "2016-12-31",
        "currently_active": false
        // location trimmed
      },
      {
        "degree_title": "BSc Computer Engineering",
        "school_name": "University of Lagos",
        "degree_type": "Level 6",
        "start_date": "2010-01-01",
        "end_date": "2014-12-31"
        // description, currently_active and location trimmed
      }
    ],
    "work_experiences": [
      {
        "company_name": "Deliveroo",
        "job_title": "Senior Backend Engineer",
        "description": "Led the migration of the order-dispatch service from a Python monolith to Go microservices on Kubernetes (Amazon EKS), cutting p99 latency by 38%. Designed an event-driven architecture based on Apache Kafka processing 40k events per second at peak. Mentored 4 engineers and ran the backend interview loop for the logistics tribe. Introduced contract testing with Pact and reduced integration incidents by half.",
        "contract_type": null,
        "start_date": "2022-01-01",
        "end_date": null,
        "currently_active": true,
        "work_experience_country": "United Kingdom",
        "work_experience_country_code": "GB",
        "work_experience_city": "London",
        "work_experience_postal_code": null,
        "experience_duration": 57
      },
      {
        "company_name": "Deliveroo",
        "job_title": "Backend Engineer",
        "contract_type": null,
        "start_date": "2019-06-01",
        "end_date": "2021-12-31",
        "currently_active": false,
        "work_experience_city": "London",
        "experience_duration": 31
        // description, country and postal code trimmed
      },
      {
        "company_name": "Ocado Technology",
        "job_title": "Software Engineer",
        "contract_type": null,
        "start_date": "2016-09-01",
        "end_date": "2019-05-31",
        "currently_active": false,
        "work_experience_city": "Hatfield",
        "experience_duration": 32
        // description, country and postal code trimmed
      },
      {
        "company_name": "Self-employed",
        "job_title": "Freelance Web Developer",
        "contract_type": null,
        "start_date": "2014-01-01",
        "end_date": "2016-12-31",
        "currently_active": false,
        "work_experience_city": "Lagos",
        "experience_duration": 36
        // description, country and postal code trimmed
      }
    ],
    "rome_jobs": [
      {
        "job_code": "200151",
        "rome_code": "M1855",
        "prediction_score": 0.846
        // job_title and rome_title trimmed
      }
      // 3 more ROME suggestions trimmed
    ],
    "languages": [
      { "language": "English", "level": "Native or Bilingual (C2)" },
      { "language": "Yoruba", "level": "Native or Bilingual (C2)" },
      { "language": "French", "level": "Limited Working Proficiency (B1)" }
    ],
    "skills": [
      { "skill_title": "Go", "skill_type": "Software skill", "status": "normalized" },
      { "skill_title": "Python", "skill_type": "Software skill", "status": "normalized" },
      { "skill_title": "Kubernetes", "skill_type": "Software skill", "status": "normalized" },
      { "skill_title": "Apache Kafka", "skill_type": "Software skill", "status": "normalized" },
      { "skill_title": "Terraform", "skill_type": "Software skill", "status": "normalized" },
      { "skill_title": "Microservices", "skill_type": "Hard skill", "status": "normalized" }
      // domain and subcategory trimmed; 25 more skills trimmed
    ],
    "certifications": [
      "AWS Certified Solutions Architect – Associate (2021)",
      "Certified Kubernetes Application Developer – CKAD (2023)"
    ],
    "interests": []
  },
  "warnings": [],
  "credits_used": 1
}

Questions about the benchmark

Which resume parser was the most accurate in this test?

HireLayer had the best results overall: 38 of 41 positions fully correct, a tie with Textkernel’s optional LLM Parser, and 26 of 28 language levels, where the next best was Affinda with 12.

Which resume parser returns language proficiency levels?

HireLayer returned a CEFR level for all 28 languages and got 26 right, including mother tongues and levels written as scores (IELTS, DELF) or Europass grids. Affinda returned 12 right levels and 7 wrong ones. Textkernel and RChilli returned the language name only.

Which resume parser handles scans and photos of a resume?

On the scanned resume and the phone photo, HireLayer returned the contact details, every position and the diplomas. Textkernel misread or cut the email on both by default, Affinda lost the phone number and garbled the job titles on the photo, and RChilli returned no work history from the scan.

Are the test resumes real candidates?

No. We wrote all ten for fictional people, so the right answer to every field is known in advance and no personal data is involved. They are free to download from this page.

Can I run this benchmark myself?

Yes. Download the ten files, run them through each parser and compare the JSON with what the resumes say. To test HireLayer, use the live demo or the free plan, which includes monthly credits and returns the same JSON as on this page.

Can I test HireLayer on my own resumes?

Yes, and it is the best test. Pick resumes that look like your real intake, write down the values you filter on, and compare. You can do it alone with the free plan, or book a demo to go through the results with us.

The best benchmark is your own resumes

Bring a sample of the CVs you actually receive and we'll go through the parsed results with you. Or start on your own with 50 free credits a month, no card required.