Cloud Natural Language API is Google Cloud's text-understanding service, exposing seven endpoints. analyzeSentiment returns document-level and sentence-level scores (score from -1.0 to +1.0 plus magnitude). analyzeEntities identifies ten entity types (PERSON, ORGANIZATION, LOCATION, EVENT, WORK_OF_ART, CONSUMER_GOOD, and more) and resolves Wikipedia links where available. analyzeEntitySentiment combines the two, attaching sentiment to each entity mention. analyzeSyntax returns dependency parses, POS tags, and lemma forms. classifyText assigns one of 700+ content categories (a v2 long-form model is available for documents that exceed the v1 length limit). moderateText flags 16 harm categories such as toxicity, insult, violence, and sexual content for UGC moderation pipelines. annotateText returns multiple features in one request but bills as if each were called separately. Billing is character-based: standard endpoints round to the nearest 1,000 Unicode characters per request (moderateText uses 100), with whitespace and HTML markup counted. The free tier covers the first 5,000 units per month for standard features, 30,000 for classifyText, and 50,000 for moderateText. Volume tiers cut cost notably at scale: sentiment runs $0.0010 per 1,000 chars from 5K to 1M units, then $0.000500 from 1M to 5M, then $0.000250 above 5M. Default project quotas cap at 600 requests per minute and 800,000 per day, with a 1,000,000-byte per-request ceiling. Higher-throughput workloads require a Cloud Quotas adjustment. Authentication uses Google Cloud's standard stack: Application Default Credentials, service account JSON keys, or OAuth bearer tokens. Eight official client libraries are maintained in separate googleapis/{lang} repos: C++, C#, Go, Java, Node.js, PHP, Python, Ruby. The caveat for cost planning is that short-text high-frequency workloads pay the per-1K-char rounding penalty. A 200-char API ping still bills one full unit, so batch where you can.