Tony Dinh drops a $124 i18n SaaS for 75 cents of tokens
He says the home-built version is more accurate, because it knows which screen the string lands on.
Tony Dinh says he has replaced another SaaS, this time the tool he was paying $124 a month for internationalization. In its place his app now runs a built-in translation harness. He puts the job at 2,705 translation keys across 16 locales, at a cost of roughly $0.75 using GPT-Luna 6.
My app now uses a built-in translation harness to do much more accurate AND in-context translations.
2,705 translation keys x 16 locales. Costs ~$0.75 with GPT-Luna 6
The money is the easy part of that post, and it is also the least interesting part. The claim Dinh leads with is accuracy. He says the in-house version produces "much more accurate AND in-context translations", and he spells out what he means by in-context. The translator knows about the app. It has a list of internal words. It knows where a given string appears, which feature and which screen. It knows how much screen space is available for the text and whether it can wrap.
Why that matters more than the price
A translation vendor sees a key and a string. It does not see your product. Every awkward translation most founders have shipped comes from that gap, a button label that runs three words long in German, a term of art from your own app rendered as its dictionary meaning. Dinh's harness closes the gap by handing the model the things only the app knows.
That is the pattern worth noticing here, and it is not really about saving $124 a month. The build-versus-buy argument used to hinge on cost and maintenance. The argument Dinh is making is that the thing he built is better at the job, because he can feed it context a third party never had access to. The vendor could not know your screen widths. You do.
One caveat, and it is the usual one. This is one founder reporting his own numbers on his own app, not a benchmark, and "more accurate" is his judgement of his own output rather than anything measured here. His verdict on the effort involved was short.
"Seriously, it's so easy."
Steal the numbers, not the vibes
One email a day. Under three minutes. Real MRR, real pricing, real launches from founders building with AI.