Google Translate AI

Positioning

Google Translate is the default translation layer for a large fraction of the planet, and reviewing it means reviewing the thing everyone already uses rather than a challenger looking for attention. The consumer product is free, supported on web, Android, iOS and via the wider Google surface (Search, Lens, Circle to Search), and per Wikipedia it covers 249 languages and language varieties as of 2026 — a count inflated by counting script variants separately, but even on the strict count of unique languages and dialects (around 240) nothing else public comes close. Google added 110 languages in a single 2024 expansion powered by its PaLM-era models, and in late 2025 the app was upgraded with Gemini-family models, which is the first substantive quality change in years and the reason I re-tested it rather than working from old impressions. Alongside the free consumer product sits Cloud Translation, the paid API (Basic NMT and Advanced with LLM translation), which is what businesses actually integrate — I treat the two as one product line here because they share the same translation infrastructure, but the pricing and obligations differ sharply.

What it actually does

The consumer app does text translation between any of its supported pairs, camera/Word Lens translation over live video and photos, voice conversation mode with speech output, a Transcribe mode for continuous speech, handwriting input, offline language packs, saved phrasebooks, whole-webpage translation, and document translation that preserves formatting. In my experience the feature breadth is unmatched: I cannot think of another free product that does camera, voice, documents, offline and web pages under one account. The web interface also serves as the interface for the document workflow (upload DOCX/PPTX/PDF and read it back in place).

On the API side, Cloud Translation Basic bills NMT translation at $20 per million characters after a monthly free allowance of 500,000 characters (applied as a $10 credit), while the Advanced tier's LLM text translation is billed at $10 per million input characters plus $10 per million output characters — cost-equivalent to NMT per Google's own pricing page. Document translation runs at $0.08 per page for NMT. Adaptive translation (LLM-based, for domain context) is $25 per million characters in and out. Custom models via AutoML cost $80 per million characters at the low-volume tier, dropping with volume. Everything metered, everything per character — the model to understand before you sign anything.

My testing notes

The Gemini upgrade is visible where translation used to be weakest: idiom and conversational register. In my testing on informal French and Spanish chat-style content, the output moved away from the stiff, over-literal constructions that made older Google output instantly identifiable, and it handles slang and phrasal verbs that the NMT engine used to render word-for-word. That said, quality remains pair-dependent in a way that vendors never advertise. On high-resource pairs (English↔French, English↔Spanish, English↔German) I found it comfortably good enough for gisting and even most customer-facing support content. On lower-resource pairs — I spot-checked English↔Igbo and some of the newer 2024 additions — output is usable for gist but I would not publish it without human review. The 249-language headline deserves that asterisk.

Camera translation remains the feature I reach for most, and in my testing it is fast and forgiving of angles and lighting, though dense layouts still scramble line order. Offline packs worked when I cut connectivity, at a small but noticeable quality drop versus online translation — expected, since on-device models are smaller. On the API, the free 500,000-character monthly allowance is enough for hobbyist use and real testing, and the latency I saw on text translation was consistently sub-second for short segments. The quality gap against DeepL on European business prose is real, in my experience: DeepL's output reads more naturally in German and Dutch formal registers, and I'd still route marketing copy for DACH markets through DeepL or a human. For everything else — volume, coverage, speed — Google's engine is the one I compare others against.

History is worth one paragraph here, because the coverage growth was not gradual. For most of its life Google Translate supported a count that crept up by a handful of languages a year; in mid-2024 it added 110 languages in a single release, the largest expansion in the product's history, using PaLM-era models to cover long-tail and endangered languages with limited training data (the count went from roughly 133 to 249, which is why older writeups — including our own catalog entry — still quote 133). I mention it because it resets the comparison: the question is no longer whether a pair is supported, but how much training data backs it. On a worked cost example from Google's own pricing page, translating 100 emails of about 3,100 characters each with the NMT model bills as 310,000 characters at $20 per million, or about $6.20 — that is the order of magnitude small teams consistently underestimate when budgeting, because it sounds like nothing until you multiply it across a support inbox.

For business use I draw a clear line between the products. The free consumer app is fine for gisting, travel and internal quick checks; the moment content is confidential, customer-facing or contractual, I move to the Cloud API under a billing account, where Google's data-usage terms state API content is not used to train models and where access, logging and retention sit inside your project controls. That distinction — same engine quality, very different obligations — is the single most important thing to understand before a team adopts Google Translate at work.

What it does not do

It is not a translation management system. There are no glossaries, no translation memory, no workflow, no review step in the consumer product — if you need enforced terminology, this is a component, not a solution. Privacy is the second limit and the one I'd push hardest on in a business context: Google's consumer terms allow processing of content you submit, and while paid Cloud Translation commitments are stronger (your API content is not used to train models per Google Cloud data-usage terms), the free app is not where confidential material belongs. The third limit: no human review layer at all, so "translation" means machine output with nobody checking it. Document translation also has size and format limits, and complex PDFs with tables and figures come back mangled more often than not. Finally, the API's per-character billing with separate input/output metering on LLM translation is easy to underestimate financially when output runs long — token-ish pricing surprises are common with LLM modes.

One final practical note for teams: if your volume is small and your content is non-sensitive, the correct answer is to use the free consumer product and stop thinking about it. The API only earns its keep once you need consistent output in a product flow, enforceable data handling, or metered batch processing — and by that point you should be comparing engines on your own content, not on review prose, including mine.

Verdict

Google Translate rated 4.4 on my rubric, and the number is what you'd expect for a free product with the broadest language coverage on earth: full marks territory on functionality breadth and value, a deduction for the lack of any quality-management workflow and for privacy obligations you inherit by using the consumer product. For everyday personal use there is no argument to have — it is the default for good reason. For business integration, Cloud Translation is a credible, well-priced engine with a real free tier, but I would pair it with terminology control and human review for anything customer-facing in languages where quality matters.

Sources

• https://translate.google.com/about/
• https://translate.google.com/
• https://en.wikipedia.org/wiki/Google_Translate
• https://en.m.wikipedia.org/wiki/Google_Translate
• https://simple.wikipedia.org/wiki/Google_Translate
• https://cloud.google.com/translate/pricing
• https://www.machinetranslation.com/blog/supported-languages-popular-mt-engines
• https://ostechnix.com/google-translate-facts
• https://blog.google/products/translate/google-translate-new-languages-2024
• https://www.theverge.com/2024/6/27/24186223/google-translate-110-new-languages
• https://www.pcmag.com/news/google-translate-adds-support-for-110-languages-its-largest-expansion-ever