VS HUMAN PROFESSIONALS

AI vs Human Translation: Tested Against Professionals (2026)

FAMILIAR RESEARCHALL ARTICLES

ABSTRACT

Familiar matches first-pass expert professional translators: 100.3% of their quality, averaged across French, Chinese, Hindi, and Indonesian, scored blind against expert reference edits from WMT24++ (August 2026). Familiar is also 5× cheaper than professional human translators on the words alone.
100.3%

of expert human translation quality

FRENCH, CHINESE, HINDI, INDONESIAN · FIRST PASS FOR EXPERT PROFESSIONAL TRANSLATORS
5×

cheaper than professional human translators

VS $22–45 PER SPOKEN MINUTE, TRANSLATION ALONE · SAME QUALITY (FIRST PASS)
ALL-IN

one Familiar render: translation, the speaker's voice, scene audio, and lip-sync

TRANSLATION ALONE IS WHAT THE PROFESSIONAL RATE BUYS

01.The test

Each source segment was translated twice, once by our production pipeline and once by a professional human translator; both versions were judged blind against expert reference edits.

  • The score: each language's result is the pipeline's relative score against the first-pass professional, with the professional fixed at 100%.
  • The human side is not ours: the professional translations and expert reference edits come from WMT24++ (paper), the public evaluation set extending WMT24 to 55 languages, produced and post-edited by professional linguists.
  • "First pass": the professional's first deliverable, before a second linguist's revision round; it is what published per-word rates buy.
THE TEST SET · WMT24++ (GOOGLE) · DEUTSCH ET AL. · ARXIV:2502.12404
Top of the WMT24++ paper: the title, the Google and Unbabel author list, and the abstract

02.Results by language

FIG. 01 · RELATIVE SCORE VS THE FIRST-PASS EXPERT PROFESSIONAL · DASHED LINE = THE PROFESSIONAL'S 100%
French100.0%
Chinese98.2%
Hindi103.3%
Indonesian99.6%
PROFESSIONAL = 100% · 95% RANGES IN THE TABLE BELOW
RELATIVE SCORE AND 95% RANGE, PER LANGUAGE · PROFESSIONAL = 100%
LanguageRelative score95% range
French100.0%97.0–105.4
Chinese98.2%95.6–100.0
Hindi103.3%100.4–108.5
Indonesian99.6%97.6–103.2
Average100.3%—

Read the ranges, not the decimals: a range that spans 100 is a statistical tie with the professional. Averaged, 100.3% is a match, not a win.

03.The price of that minute

  • Published professional rates run $0.15 to $0.30 per word; at about 150 spoken words a minute, $22 to $45 per spoken minute, text only.
  • Familiar is priced by Studio contract; one render is all-in: translation, the speaker's voice, scene audio preserved, and the lip-sync.

04.The whole job, at published prices

Full human dubbing, translation plus voice, runs $70 to $150 per finished minute. The per-word rate is only the first line item; the published rate cards for the work around it:

PRICE CONTEXT IN US DOLLARS, PER TARGET LANGUAGE
OptionPer finished minuteFor 60 minutes
Published human translation + review$10 median
$11.95 mean		$600 median
$717 mean
Published human dubbing packages (voice, direction, mix)$39–$94
~$60 midpoint		$2,340–$5,640
~$3,600 midpoint
Full human dubbing, translation plus voice$70–$150$4,200–$9,000
Premium studio dubbingFrom $128From $7,680
FIG. 02 · A 60-MINUTE VIDEO INTO ONE LANGUAGE · MEDIANS AND RANGE MIDPOINTS
TRANSLATION$600
VOICE PACKAGE~$3,600
FULL HUMAN DUB~$6,600

05.The same translation, inside a finished dub

This study scores translation text alone; the full head-to-head on finished dubs from the same production pipeline is Familiar vs ElevenLabs Dubbing: Two Paired Studies.

54.7%

fewer important translation errors than ElevenLabs' newest Dubbing v2 (Alpha)

29 VS 64 · 111 PAIRED DUBS · MANDARIN, SPANISH, JAPANESE

06.Where humans still win

A professional brings what no metric captures: transcreation of wordplay, legal and brand review, accountability for a single high-stakes document. For spoken video at catalogue scale, the equation flips.

  • Do Not Translate list: names and catchphrases stay exactly as spoken.
  • Review screen: every video pauses before render; any line can be edited.

To test the translation on your own footage: How to Run an AI Dubbing Pilot (2026).

REFERENCES

  1. [1]Familiar vs ElevenLabs benchmark: full results, intervals, and listening examples
  2. [2]WMT24++ dataset: the professional translations and expert reference edits
  3. [3]WMT24++: Expanding the Language Coverage of WMT24 to 55 Languages & Dialects (paper)
  4. [4]Netflix Post Production: published translation rate card
  5. [5]Voquent: how much does dubbing cost?
  6. [6]Gotham Lab: dubbing services (premium studio rates)
  7. [7]ATA 2025 Compensation Survey (interpreter rates)
  8. [8]Smartling: Translation rates -- how much does translation cost?
  9. [9]Tomedes: Translation rates and why translation prices vary

Dubbing is finally good. See the measurements, then talk with the team about your catalog.

See the full benchmark →Talk with the team

FAMILIAR · THE LAUNCH FILM · 2:31