Choose the right model for your application

Cloud Translation - Basic API gives you straightforward access to Google's Neural Machine Translation (NMT) model without any modification, while Cloud Translation - Advanced lets you choose between the default NMT model, a state-of-the-art LLM translation model that supports Adaptive Translation, and your own customized versions of the TLLM or NMT models.

Use this table to decide on the best model for your use case.

Model Description Model ID
Translation LLM (TLLM)

Highest quality translation model - a Large language model (LLM) powered by Gemini and fine-tuned for translation to provide the highest quality translations at fast latencies (2x faster latency than Gemini 2.0 Flash).

general/translation-llm
Neural Machine Translation (NMT) model

Fastest translation model - ideal for real time and latency critical use cases. This is an updated version of Google's original Neural Machine Translation, available through the Cloud Translation - Basic API using [v2 client libraries](/translate/docs/reference/libraries/v2/overview-v2). You can get going quickly using this model out of the box, without any customization.

general/nmt
Customized Translation LLM (Preview)

Provides high quality translation customization via fine-tuning of the Translation LLM with domain-specific language-pair training data.

This is a new feature of the Translation LLM model.

translation-llm-custom/{model-id}
Adaptive Machine Translation model

This option involves real-time, lightweight customization of the Translation LLM using Adaptive Translation based on a limited dataset of translation examples.

general/translation-llm-adaptive
Customized NMT model You can train a new customized version of the NMT model using training data specific to your domain and use case. The model ID is generated after model training completes.

Note that each model supports its own set of languages. Refer to the Supported languages guide to ensure that your source and target languages are supported.