Google Cloud offers translation models for quality and speed

Google Cloud’s Gemini Enterprise Agent Platform supports Translation LLM and Neural Machine Translation (NMT) for text translation through its console and APIs. Translation LLM is powered by Gemini and fine-tuned for translation, prioritizing natural phrasing over literal rendering. Google says it delivers higher translation quality than NMT, while NMT targets real-time applications with ~100ms latency.

Translation LLM supports advanced glossaries, supervised fine-tuning and adaptive translation using example sentence pairs to match a customer’s style, tone or industry domain. In Vertex AI Studio, developers can upload examples in TMX or TSV files. API requests support up to 5 sentence pairs inline. Cloud Translation API users can also create persistent datasets of examples and reuse them across requests.

Both models support HTML translation, while NMT also supports batch and formatted document translation. Supported source and target languages vary by model and should be checked before use. For customized translations in Vertex AI Studio, reference examples count toward the input limit of 3,000 characters per request. Output is also limited to 3,000 characters, with excess text dropped.