Google Translate

Google Cloud Translation offers the widest language coverage among translation APIs — 130+ languages including many that DeepL doesn't support. Two API versions exist: Basic (v2, simpler) and Advanced (v3, full-featured). For agents, the Advanced API provides glossaries, batch translation, AutoML custom model support, and document translation. The Basic API is sufficient for straightforward text translation with automatic language detection. Google's neural machine translation quality is strong globally, with particular strength in Asian languages (Chinese, Japanese, Korean) and less-common language pairs where DeepL has limited support. The GCP auth model (service accounts, OAuth) adds integration complexity compared to DeepL's simple API key. For agents needing broad language support or GCP ecosystem integration, Google Translate is the standard choice.

Authentication uses GCP's standard model: service accounts with JSON key files, or OAuth 2.0 for user context. For agents, service account authentication is the recommended path. The GOOGLE_APPLICATION_CREDENTIALS environment variable points to the service account key file. IAM roles control access: Cloud Translation API User for basic usage, Cloud Translation API Editor for glossary management. The GCP auth model is more complex than API key-based translation services but provides stronger access control. API keys are supported for the Basic v2 API (simpler but less secure). The Advanced v3 API requires service account or OAuth auth. For agents already in the GCP ecosystem, the auth model is familiar. For agents outside GCP, the service account setup adds onboarding friction.

Documentation at cloud.google.com/translate/docs is comprehensive, following GCP's standard documentation pattern. Quickstart guides, API reference, client library documentation, and pricing information are well-organized. Client libraries are available for Python, Node.js, Java, Go, C#, PHP, and Ruby — all auto-generated from the API spec and well-maintained. The documentation distinguishes clearly between Basic and Advanced APIs. REST and gRPC APIs are both documented. Code samples are provided for each language. The documentation's main weakness: navigating between Basic and Advanced API docs requires attention to version markers. GCP's documentation ecosystem (Codelabs, YouTube, Stack Overflow tags) provides extensive supplementary learning resources. For agents, the Python and Node.js client library docs are the fastest path to integration.

Basic API (v2): POST to /language/translate/v2 with q (text array), target, and optional source. Simple and effective for text translation. Language detection: GET /v2/detect. Advanced API (v3): POST to /v3/projects/{project}/locations/{location}:translateText with contents array, targetLanguageCode, and optional glossaryConfig. Batch translation processes large document sets asynchronously via Cloud Storage. Document translation handles PDF, DOCX, and other formats. Glossary management creates and applies terminology mappings. The v3 API uses GCP's standard resource naming conventions (projects/locations). AutoML models enable custom training on domain-specific parallel corpora. The API supports HTML and plain text input with format preservation. Response payloads include translations and detected source languages.

API errors follow GCP's standard error format with code, message, and status fields. Quota limits are per-project with default allocations (characters per minute, per day). Quota increases are available via GCP console. 429 errors indicate quota exhaustion. The Basic v2 API has simpler quotas. The Advanced v3 API quotas cover characters, documents, glossary operations, and batch translation separately. Batch translation jobs are async with operation polling for completion. Language detection confidence scores help agents decide whether to trust detected language. The main reliability consideration: GCP's quota model requires agents to understand and monitor multiple quota dimensions. The free tier (500K characters/month for Basic, separate for Advanced) provides adequate development allowance.