The Rare-Language Premium Vanished. The Queues Came Back at $65.

2026-08-27 · Expert Match AI team

Eight weeks ago we published a post arguing that bilingual work is the AI training niche with the least competition. The centrepiece was a counterintuitive finding straight from live listings: the world's *biggest* languages paid the least, because that is where the applicants are. French, Hindi and Hebrew sat at $10–30/hr, while Tamang and Tibetan language experts were listed at $40–95/hr, Tongan at $45–95, and Dioula up to $95.

We have gone back twice since to check. Here is what happened to that market.

All four rare-language roles are gone. So are French, Hindi and Hebrew as originally listed. Of the fifteen language roles that post cited, not one is still on the board.

But the field did not stay collapsed. It rebuilt - into something with a very different shape.

The three snapshots

| | July | Aug 11 | Aug 27 | |---|---|---|---| | Linguistics listings | 27 | 12 | 18 | | Distinct languages named | dozens | 14 | 15 | | Top language-role rate | $95/hr | $65/hr | $65/hr | | Field average | $46/hr | $55/hr | $58/hr |

The count has recovered two-thirds of the way back. The ceiling has not moved at all. That is the whole story in two rows: the volume returned and the premium did not.

The board today, in full

Every linguistics listing we currently track:

| Role | Platform | Rate | |---|---|---| | Surge AI - Journalists & Professional Writers | Surge | $200–$400/hr | | Indonesian Bilingual Expert | micro1 | $60–$65/hr | | English Voice Coach | micro1 | $40–$65/hr | | Bengali Bilingual Expert | micro1 | $30–$65/hr | | Dutch Bilingual Expert | micro1 | $30–$65/hr | | Hebrew Bilingual Expert | micro1 | $30–$65/hr | | Italian Bilingual Expert | micro1 | $30–$65/hr | | Polish Bilingual Expert | micro1 | $30–$65/hr | | Russian Bilingual Expert | micro1 | $30–$65/hr | | Tamil Bilingual Expert | micro1 | $30–$65/hr | | Thai Bilingual Expert | micro1 | $30–$65/hr | | Turkish Bilingual Expert | micro1 | $30–$65/hr | | Vietnamese Bilingual Expert | micro1 | $30–$65/hr | | Voice Recorder (French + English) | micro1 | $30–$65/hr | | Voice Recorder (Japanese + English) | micro1 | $30–$65/hr | | Voice Recorder (Korean + English) | micro1 | $20–$65/hr | | Outlier - Expert Writing & Editing | Outlier | $15–$35/hr | | Outlier - Languages & Localization | Outlier | $15–$30/hr | | English Linguistics | Braintrust | $10–$30/hr |

Look at the middle of that table. Eleven separate micro1 listings, eleven different languages, and the identical band $30–65/hr on every one of them. Bengali and Dutch pay the same. Tamil and Italian pay the same. A language spoken by 240 million people and one spoken by 24 million are priced identically.

That is not a market discovering scarcity. That is a rate card.

Fifteen languages appear in the data: Indonesian, Bengali, Dutch, Hebrew, Italian, Polish, Russian, Tamil, Thai, Turkish, Vietnamese, French, Japanese, Korean and English. Every one is a major world language. The Himalayan, Pacific and West African listings that made July's argument are simply not there any more.

(A filing note: the Thai listing is categorised under ML, Data & AI rather than Linguistics on our own board - so a reader filtering by Linguistics would miss it. We are flagging it rather than quietly fixing the count.)

Why the premium disappeared

Language queues are the most project-shaped work on this board, and project-shaped work closes.

A lab needing Tamang coverage needs a specific quantity of Tamang data. It posts a role at a premium because almost nobody speaks Tamang and it needs to find them fast. Then it hires four people, they produce the dataset, and the queue closes - permanently, because the lab does not need a second Tamang dataset next quarter. The $95/hr was never a standing rate for a rare language. It was a finder's premium on a fixed-size job, and it expired with the job.

The rebuild tells the other half of the story. What came back was not more rare-language work at a premium; it was a standardised block of major-language listings at a fixed band, posted by a single platform. Once a lab knows roughly what a bilingual evaluation hour is worth, it stops bidding for scarcity and starts publishing a price.

Compare that to the attorney roles which have climbed to $140–400/hr over the same period. Legal evaluation is ongoing: every model version needs re-evaluating, every jurisdiction is different, the work renews. Language data collection frequently does not.

This is the clearest example we have yet found of a dynamic that runs through the whole market, and it is worth internalising: a high listed rate tells you about urgency, not about durability.

What was right and what was wrong

Right: competition in rare-language queues genuinely is thin, the premium genuinely was four-fold, and the post's closing advice - "check back weekly, new languages appear constantly, the first qualified applicants to a fresh queue face the least competition" - turns out to be the single most important sentence in it. That is precisely how this niche has to be worked. A Tibetan listing was live on our board again as recently as this month, and is gone again now. Weekly is not an exaggeration.

Wrong, or at least misleading: the framing implied a standing market you could plan around. It is not one. If you read that post in July, decided to pursue AI training work because you speak Tongan, and are only getting to it now, the door you were aiming at has closed. We should have been clearer that a listing with four seats and a finder's premium is an event, not a career.

One figure also needs attributing properly. The post said micro1 runs "the majority of the 27 live linguistics roles, across 40+ languages." The 27 was our board. The 40+ was micro1's description of its own catalogue, not something we counted, and it sat close enough to our own computed numbers to read as though we had verified it. We had not. Our board has never shown more than 27 language listings at once, and shows 15 languages today.

So is the niche still worth working?

As a standing income, no - not at $10–65/hr on a board where the entire major-language tier is priced identically.

As an opportunistic play, yes, and the case is now clearer than it was. The mechanism that produced $95/hr for Dioula speakers has not gone away; it fires, completes, and fires again the next time a lab needs coverage in a language nobody has collected. What has changed is how you have to work it: monitor rather than plan. Set a weekly check on the linguistics board, apply within days of a rare-language listing appearing, and treat it as a windfall rather than a plan.

There is also a floor worth noticing. Indonesian is the one language on the standardised tier posted at $60–65 rather than $30–65 - double the floor for the same ceiling. When a platform breaks its own rate card for one language, that is where the unmet demand is.

And keep your primary effort where the durable money is. The one listing on this table paying real money - Surge's journalist tier at $200–400/hr - pays for editorial judgment in English, not for bilingualism. That is the general pattern: fluency alone is not scarce, but fluency plus a credential is. A bilingual lawyer should be applying to the legal roles at $140–400, not to a language queue at $30.

Methodology and corrections policy

Counts cover every live listing naming a specific language in its title, description or skills, computed against the live board on 2026-08-27 (313 roles tracked, 209 disclosing an hourly range). July figures are from the bilingual post published 2026-07-22 and the July 31 monthly report; August 11 figures are from our own mid-month pull. Roles are marked gone when they no longer appear in our pipeline; platforms occasionally unlist and relist, so "gone" means not currently public rather than permanently closed.

We leave incorrect published figures in place and correct them in a new post rather than editing silently. If you think a number here is wrong, write to support@expertmatchai.expert.

Published by the Expert Match AI team. Pay figures are listed rates from live postings, not a survey of accepted offers. Some outbound application links on this site carry disclosed referral codes; rankings and recommendations are never influenced by them.