Salesforce exposes multiple APIs because they serve different patterns.
SOAP API — XML-based, Synchronous, request-response. Each call sends/receives XML messages. Strongly typed via WSDL. Heavier than REST. Use cases: legacy integrations that already speak SOAP, integrations needing strict schema. Most modern projects pick REST instead.
REST API — JSON-based, Synchronous, lightweight. Standard HTTP verbs (GET/POST/PATCH/DELETE). Use cases: most modern integrations — web apps, mobile apps, server-to-server, low-volume batch. Default choice unless you need volume or async.
Bulk API — Asynchronous, batch-oriented, optimised for large data volumes. You submit a job containing CSV-formatted records (or JSON in Bulk API 2.0), Salesforce processes it asynchronously, you poll for results. Two versions: 1.0 (older, more configurable) and 2.0 (simpler, more automated). Use cases: any data load above ~5,000 rows. Loading 50k Account updates? Bulk API. Doing it via REST one-by-one will hit your daily call limit and take hours.
Streaming API — Push-based, server-sends-to-client. Salesforce publishes events; clients subscribe via long-polling or CometD/Bayeux. Use cases: real-time notification of record changes (Change Data Capture / Platform Events), server-to-server pub-sub. Replaces polling-for-changes patterns.
Other API flavours worth knowing:
- Composite API — bundle multiple REST calls into one HTTP request to save round-trips and stay under API limits.
- GraphQL API — newer; flexible queries with a single endpoint. Some admins prefer it for nuanced data fetching.
- Tooling API — for metadata operations (deploying classes, querying setup objects). Not for data.
- Metadata API — bulk metadata deploys, full org backups, retrieval of customisations.
- UI API — for building custom UIs that mirror Salesforce's own; respects sharing, FLS, labels, etc., out of the box.
Decision tree:
- One record, real-time, user-driven -> REST.
- Thousands of records, async -> Bulk API.
- Listening for changes -> Streaming (CDC or Platform Events).
- Metadata ops -> Metadata API.
- Building UI on top -> UI API.
- Aggregating multiple REST calls -> Composite API.
Pick based on volume, latency, and data-vs-metadata.