Agent inference

Anthropic-compatible API

What it is

Source: Anthropic, Messages API reference, read 2026-09-25

Source: RunInfra, Anthropic Messages API, read 2026-09-25

Why it moves cost and latency

Coding agents depend on tool calls and streamed events as well as generated text. Checking the provider's supported fields avoids treating an endpoint change as a promise that every Anthropic feature is available.

What it looks like in practice

RunInfra accepts POST https://api.runinfra.ai/v1/messages with x-api-key or Authorization: Bearer authentication, tools with JSON Schema input, stream: true for Server-Sent Events, and thinking.type set to enabled, adaptive, or disabled. The cache_control field and anthropic-beta header are accepted but ignored. Anthropic server tools, including bash, text editor, web search, and computer use, and Anthropic-only document blocks are rejected.

Related terms

Questions this definition answers

What does Anthropic-compatible API mean in inference serving?

An Anthropic-compatible API accepts requests in the Anthropic Messages format so clients can use another model provider. Compatibility covers the supported request, response, and streaming shapes; it does not imply full Anthropic feature parity. RunInfra accepts POST https://api.runinfra.ai/v1/messages with x-api-key or Authorization: Bearer authentication, tools with JSON Schema input, stream: true for Server-Sent Events, and thinking.type set to enabled, adaptive, or disabled. The cache_control field and anthropic-beta header are accepted but ignored. Anthropic server tools, including bash, text editor, web search, and computer use, and Anthropic-only document blocks are rejected.