Claude Native /v1/messages API — Caching and Thinking

Call Claude via POST /v1/messages using the Anthropic SDK and KiosAPI base URL. Supports prompt caching, PDF analysis, extended thinking, and streaming.

KiosAPI supports Claude’s native /v1/messages endpoint format — use it directly with the Anthropic SDK to get full access to prompt caching and extended thinking. Prompt caching lets you mark large, repeated blocks of context (such as lengthy PDFs or system prompts) so they are cached server-side and reused across requests, significantly reducing both latency and cost.

The native format only lists commonly used parameters. For the complete parameter reference, see the Claude official documentation.

Supported Models

Claude native format supports models with the cld- or claude- prefix. It also supports kimi-k2-0711-preview, qwen3-coder-plus, and glm-4.5+.

Request

POST https://kiosapi.com/v1/messages

Headers

Header	Required	Description
Content-Type	Yes	application/json
x-api-key	Yes	sk-xxx
anthropic-version	Yes	2023-06-01

Request Body

model · string · required

Claude model name (e.g. claude-sonnet-4-6).

messages · array · required

Array of message objects forming the conversation.

Show messages properties

messages[].role · string · required

Role of the message author: user or assistant.

messages[].content · string | array · required

The message content. Pass a plain string for text-only messages, or an array of content objects for multimodal input such as PDFs and images.

max_tokens · number

Maximum number of tokens to generate in the response.

temperature · number

Sampling temperature between 0 and 2.

top_p · number

Nucleus sampling. Use as an alternative to temperature.

top_k · integer

Only sample from the top K options for each subsequent token.

stream · boolean

Set to true to stream the response using server-sent events.

tools · object

Function calling tools available to the model.

thinking · object

Extended thinking configuration. Available on claude-3-7-sonnet only.

Content Types for PDF and Image Analysis

When passing content as an array, each element uses one of the following structures:

Type	Sub-type	Fields	Description
image or document	base64	source.type: "base64", source.data, source.media_type	Base64-encoded file
image or document	url	source.type: "url", source.url	File accessible by URL
text	—	text	Plain text prompt

Cache Control

To cache a file or block of content across multiple requests, add a cache_control field to the content object. This is especially useful for large PDFs or system prompts you send repeatedly:

{
  "type": "document",
  "source": {
    "type": "base64",
    "data": "JVBERi0xLjQK...",
    "media_type": "application/pdf"
  },
  "cache_control": {
    "type": "ephemeral"
  }
}

Examples

cURL (Chat)

curl https://kiosapi.com/v1/messages \
  -H "Content-Type: application/json" \
  -H "x-api-key: sk-xxx" \
  -H "anthropic-version: 2023-06-01" \
  -d '{
    "model": "claude-sonnet-4-6",
    "messages": [
      {
        "role": "user",
        "content": "Hello, who are you?"
      }
    ],
    "max_tokens": 1688,
    "temperature": 0.5,
    "stream": false
  }'

Python (PDF with Caching)

import anthropic

client = anthropic.Anthropic(
    api_key="sk-xxx",
    base_url="https://kiosapi.com"
)

response = client.messages.create(
    model="claude-sonnet-4-6",
    max_tokens=1000,
    messages=[
        {
            "role": "user",
            "content": [
                {
                    "type": "text",
                    "text": "Summarize this document"
                },
                {
                    "type": "document",
                    "source": {
                        "type": "url",
                        "url": "https://example.com/document.pdf"
                    },
                    "cache_control": {"type": "ephemeral"}
                }
            ]
        }
    ]
)

print(response.content[0].text)

Python (Extended Thinking)

response = client.messages.create(
    model="claude-3-7-sonnet-20250219",
    max_tokens=8000,
    thinking={
        "type": "enabled",
        "budget_tokens": 4000
    },
    messages=[{"role": "user", "content": "Solve this complex problem..."}]
)

Response

{
  "id": "msg_014AoBefsejHUjbdRntn7euw",
  "type": "message",
  "role": "assistant",
  "content": [
    {
      "type": "text",
      "text": "Hello! I'm Claude, an AI assistant. How can I help you today?"
    }
  ],
  "model": "claude-sonnet-4-6",
  "stop_reason": "end_turn",
  "stop_sequence": null,
  "usage": {
    "input_tokens": 12,
    "output_tokens": 38
  }
}

Response Fields

id · string

Unique message identifier.

type · string

Always "message".

role · string

Always "assistant".

content · array

Array of content blocks in the response.

Show content properties

content[].type · string

Content block type, e.g. "text".

content[].text · string

The text of the response.

model · string

The model that generated the response.

stop_reason · string

Reason the model stopped generating. One of end_turn, max_tokens, or tool_use.

usage · object

Token usage for the request.

Show usage properties

usage.input_tokens · integer

Number of input tokens consumed.

usage.output_tokens · integer

Number of output tokens generated.