/messages

Send and receive messages using Anthropic's API format. This endpoint is compatible with Claude and other Anthropic-style APIs.

POST

/v1/messages

Send a structured list of input messages and receive a model-generated message.

Request Body

modelThe model to use (required)

messagesArray of message objects (required)

max_tokensMaximum tokens to generate (required)

systemSystem instructions or context (optional)

streamEnable server-sent events streaming (optional)

temperatureSampling randomness control (optional)

toolsAvailable functions for tool use (optional)

tool_choiceTool invocation behavior (optional)

reasoningThinking (reasoning) control (optional): {"effort": "low"} sets the level, {"max_tokens": 2000} caps thinking tokens, {"enabled": false} turns it off. Supported models and defaults: see Reasoning Control

Request Body

{
  "model": "your-model",
  "max_tokens": 1024,
  "system": "You are a helpful assistant.",
  "messages": [
    {"role": "user", "content": "Hello!"}
  ],
  "reasoning": {"effort": "low"}
}

Anthropic's thinking parameter ({"type": "enabled", "budget_tokens": ...} / {"type": "disabled"}) has no effect on OpenRouter-routed models: on 2026-09-03 a kimi-k3 request with {"type": "disabled"} still returned a thinking block. Use the reasoning object on this endpoint as well.

Response

Returns a message object with content array, model identifier, stop reason, and token usage metrics.

Response

{
  "id": "msg_abc123",
  "type": "message",
  "role": "assistant",
  "model": "your-model",
  "content": [
    {"type": "thinking", "thinking": "The user greeted me..."},
    {"type": "text", "text": "Hello! How can I help you today?"}
  ],
  "stop_reason": "end_turn",
  "usage": {
    "input_tokens": 20,
    "output_tokens": 45
  }
}

On reasoning models the thinking comes back as a {"type": "thinking"} block in the content array ahead of the text block (thinking_delta when streaming). Those tokens are counted in usage.output_tokens and bill at the output rate.