Claude API is Anthropic's production interface to the Claude family of large language models. As of 2026, the family includes Claude Haiku 4.6 (fast and cheap), Claude Sonnet 4.6 (balanced), and Claude Opus 4.6 (most capable). Claude is known for long-context handling (200K-plus tokens), carefully calibrated reasoning, and strong tool use.
Claude's API follows the standard chat/messages format. You send a list of messages (system, user, assistant turns) and optionally a set of tool definitions. Claude returns a response that may include text and structured tool calls. For agentic workflows, the client executes the tool calls and sends results back as the next user turn, looping until the model emits a stop signal or a 'done' tool call. Streaming is supported via server-sent events for low-latency UX. Rate limits scale with account tier — brand-new accounts get ~1M tokens/minute on Sonnet, which scales up as you spend.
Haiku 4.6: $0.25 per million input tokens, $1.25 per million output tokens. Sonnet 4.6: $3 input, $15 output. Opus 4.6: $15 input, $75 output. Prompt caching can reduce input costs by up to 90% on repeated prompts. Batch API offers 50% discount on non-urgent workloads. Claude Mythos (the frontier gated model) is not available on the API — see Project Glasswing.
Notion, Quora, Zoom, Slack, Vanta, DoorDash, GitLab, Asana, Thomson Reuters, and thousands of startups. Claude Sonnet 4.6 is the default choice for most production RAG, agents, and code generation tasks in 2026.