AI training jobs for bilingual speakers: which languages pay what, and why
Pay breakdown · 2 weeks ago
Speaking a second language is the most common way into AI training work, but the same job can pay three times more in one language than another. From the 135 non-English roles on this site: which templates pay the same in every language, which pay by market, what a professional skill adds, and where each language does best.
The short version
Of the 632 open roles on this site on 26 September 2026, 135 require a language other than English. They pay anywhere from $6 to $200 an hour. Very little of that spread is about how hard the language is. It comes from three things, in this order:
- Whether the job carries a professional skill on top of the language (STEM, medicine, voice acting, law).
- Whether the platform prices the template by market or pays one rate everywhere.
- Where you live, for the minority of listings that restrict residence.
Knowing which of those applies to a listing tells you more than the language does.
Some templates pay the same in every language
Several platforms run one job template across many languages at a single rate. For these, your language decides whether you can apply, not what you earn.
- Invisible Technologies pays a flat $17 an hour across its voice and audio evaluation family (Arabic, Gulf Arabic, French, Portuguese), its dialect and localisation listings (Kansai Japanese, Cordobés Spanish, Cantonese, Hindi dialects), and $21 across its coding-and-JSON evaluation family. Its Danish and Dutch speaker projects pay $1.30 per unit instead.
- Mercor's Audiobook QA listings pay $15 to $20 whether you work in Dutch, French, German, Italian or either variety of Portuguese.
- micro1's Language Expert listings pay $20 to $36 in every language we list (French, German, Hindi, Japanese and others), and its Bilingual Expert listings $30 to $65 (Polish, Turkish, Vietnamese and others).
If your language is priced low elsewhere (see below), flat-rate templates are where it is worth the most. A Tamil or Thai speaker earns the same on micro1's bilingual template as a Dutch speaker does.
Other templates pay by market
Mercor prices several families by language, and the tiers are stark.
AI Safety Experts (English plus another language):
| Language | Pay per hour |
|---|---|
| Danish, Dutch, Finnish, Norwegian, Swedish | $48–62 |
| Portuguese (global) | $29–45 |
| Thai | $24–35 |
| Indonesian, Malay, Vietnamese | $17–25 |
| Gujarati, Kannada, Malayalam, Marathi, Odia, Tamil, Telugu, Urdu | $16–22 |
The template and task description are identical, yet the rate differs threefold: compare English and Danish against English and Tamil.
Sonic Audit Specialist follows the same pattern: French and German at $41.50, Spanish and Italian at $39.50, Japanese at $37.50, Korean and Portuguese at $25, Chinese at $21.50 and Hindi at $18.
The pattern tracks local wage levels and how scarce speakers are on the platform. Hindi and Mandarin have enormous speaker populations, which keeps their rates low; Norwegian and Finnish do not.
A professional skill is worth more than a rarer language
Mercor's bilingual AI safety family runs each language at two levels, generalist and STEM (usually a PhD). The STEM premium is consistent:
- Danish: $48–52 generalist, $61–65 STEM
- Japanese: $48–52, $68–72
- Norwegian: $58–62, $77–81
- Thai: $18–22, $24–28
Further up, the language becomes a supporting skill. Mercor's multilingual physician roles pay $170 to $190 an hour, and micro1's Bilingual Mental Health Expert $100 to $200. Voice work sits in between: Mercor's voice-cloning sessions pay $50 to $100 an hour (to $150 for Swiss German), but that is a licence of your voice, and our privacy guide covers what to check first.
Where each language does best right now
A quick reading of the current board, by language group:
- Nordic languages (Danish, Finnish, Norwegian, Swedish) and Dutch: the best-paid general language work. Mercor's market-tiered roles favour you; start there.
- French, German, Spanish, Italian, Japanese: mid-to-high on market-tiered roles, plus the widest choice of listings. Spanish and German each have more than ten roles on the site.
- Korean, Portuguese, Chinese (Mandarin and Cantonese): priced in the middle or lower on market-tiered roles, so flat-rate templates (micro1's language and bilingual experts, Invisible's families) are often the better deal.
- Indian languages (Hindi, Tamil, Telugu, Gujarati, Malayalam, Marathi, Odia, Bengali, Kannada, Urdu): the lowest market-tiered rates on the board. micro1's bilingual template at $30 to $65 (Tamil, Bengali) is worth several times Mercor's regional rates for comparable general work, so prioritise it.
- Southeast Asian languages (Thai, Indonesian, Malay, Vietnamese): similar to the Indian group, with flat-rate templates again the better value. Indonesian also has a US-dollar ChatGPT user role at $50 for residents of Indonesia.
Three things to check on any language listing
- Which variety? Many listings name a locale: Spanish (Spain) or (Latin America), French (France) or (Canada), Portuguese (Brazil) or (Portugal), Arabic (Gulf) or (global). Applying for the wrong one wastes a qualification attempt.
- Is residence required? Most language roles are open worldwide, but some are not: Mercor's voice-cloning roles are restricted to the country of the accent, and its PDF transcription roles for Indian languages are India-only. A language locale is not a residence requirement unless the listing says so.
- Is the listing internally consistent? Templated ads are sometimes edited carelessly. Invisible's French Text and Audio listing asks for native French but describes recording Japanese; we flag that on the job page. When a listing contradicts itself, confirm the working language before onboarding.
Sources
Every figure comes from the 135 roles on this site that required a non-English language on 26 September 2026, each quoting its platform's published rate. "Language" means the listing's required working language; English is excluded from the counts. Rates change often and the listing page always outranks this guide. We are not affiliated with any of these platforms and we do not process applications.
Questions
- Which languages pay the most for AI training work?
- On market-tiered listings, the Nordic languages and Dutch pay the most for general work: Mercor's AI Safety Experts template pays $48 to $62 an hour for Danish, Dutch, Finnish, Norwegian and Swedish, against $16 to $22 for most Indian languages. But a professional skill matters more than the language: bilingual physicians and mental-health experts earn $100 to $200 an hour regardless of language.
- Why does the same AI training job pay differently by language?
- Some platforms price each language by market, reflecting local wage levels and how many speakers are available, not how hard the language is. Mercor's Sonic Audit Specialist pays $41.50 for French and German but $18 for Hindi. Other templates pay one rate everywhere: Invisible Technologies pays $17 across its voice and audio evaluation family, and micro1's Bilingual Expert pays $30 to $65 in every language.
- Do I have to live in a country to do language work in its language?
- Usually not. Most language roles on this site are open worldwide. Exceptions are stated on the listing: Mercor's voice-cloning roles require residence in the country of the accent, and its PDF transcription roles for Indian languages are India-only. A named locale such as Spanish (Spain) describes the language variety, not where you must live, unless the listing says otherwise.
- Is a STEM degree worth anything in bilingual AI training work?
- Yes, consistently. On Mercor's bilingual AI safety family, the STEM (usually PhD) version of each language pays about $6 to $20 an hour more than the generalist version: Danish $48 to $52 versus $61 to $65, Japanese $48 to $52 versus $68 to $72, Norwegian $58 to $62 versus $77 to $81.
Platforms covered here
Put this into practice
Every listing shows its pay and who it is open to.