MaaS_Cl_Opus_4.8/5

Claude-opus-4.8

Basic Information

• URL： https:// genaiapi.cloudsway.net

• API Request Endpoint: POST

• Authentication : Authentication is performed via HTTP Bearer Token, and the API key needs to be included in the request header.

Parameter Name	Type	Required	Description
Content-Type	string	is	Fixed to application/json
Authorization	string	is	Bearer {your_api_key}

Explanation of Core Parameters

Parameter	Type	Required	Description
model	string	is	Model ID to be used
messages	array	is	Conversation message list, includingrole (user or assistant) and content
max_t okens	integer	is	Maximum number of tokens generated
system	string	No	System prompt, used to set the behavior and background of the assistant
temperature	number	No	Sampling temperature, where a higher value makes the output more random and a lower value makes it more deterministic, only supports a fixed value of 1
top_p	number	No	Nucleus sampling parameter, the model considers the results of tokens with top_p probability mass, only supports passing a fixed value of 0.99
top_k	integer	No	Only sample from the K tokens with the highest probability in each step, opus 4.8 does not support this
stream	boolean	No	Whether to enable streaming, default isfalse
stop_sequences	array	No	Custom stop sequences, which stop the model when generated

Request and Response Examples

Endpoint calls /v1/messages

curl --location --request POST 'https://genaiapi.cloudsway.net/{ENDPOINT}/v1/messages' \
--header 'Authorization: Bearer {Your AK}' \
--header 'Content-Type: application/json' \
--data-raw '{
    "max_tokens": 2048,
    "stream": false,
    "messages": [
        {
            "role": "user",
            "content": "Prove the Pythagorean theorem"
        }
    ],
    "thinking": {
        "type": "adaptive"
    }
}'

OpenAI Endpoint Call /chat/completions

curl --location --request POST 'https://genaiapi.cloudsway.net/v1/ai/{ENDPOINT}/chat/completions' \
--header 'Authorization: Bearer {Your AK}' \
--header 'Content-Type: application/json' \
--data-raw '{
    "max_tokens": 2048,
    "stream": false,
    "messages": [
        {
            "role": "user",
            "content": "Prove the Pythagorean theorem"
        }
    ],
    "thinking": {
        "type": "adaptive"
    }
}'

Unified domain name calls chat/completions

curl --location --request POST 'https://genaiapi.cloudsway.net/v1/chat/completions' \
--header 'Authorization: Bearer {Your AK}' \
--header 'Content-Type: application/json' \
--data-raw '{
    "model": "MaaS_Cl_Opus_4.8_20260528",
    "stream": false,
    "messages": [
        {
            "role": "user",
            "content": "who are you"
        }
    ],
    "thinking": {
        "type": "adaptive"
    }
}'

Function Introduction

thinking

opus4.8 does not support extended thinking budget, and passing the following parameters will result in a 400 error

thinking: {"type": "enabled", "budget_tokens": N}

opus4.8 only supports the Adaptive mode

thinking: {"type": "adaptive"}

Minimum cached token

The minimum cacheable prompt length on Claude Opus 4.8 is 1024 tokens, and no code changes are required.

Introduction to Claude-opus-5 Features

Mid-conversation Tool Change (Beta)

The request header must carry: mid-conversation-tool-changes-2026-07-01to dynamically add and remove tools between multiple conversation rounds while retaining the prompt cache.

{
    "model": "claude-opus-5",
    "max_tokens": 1024,
    "cache_control": {"type": "ephemeral"},
    "system": "You are a code review assistant. Be concise.",
    "messages": [
      {
        "role": "user",
        "content": "Review process() in utils.py for performance issues."
      },
      {
        "role": "assistant",
        "content": "The list comprehension is fine for small inputs. For large inputs, consider a generator to avoid materializing the full list."
      },
      {
        "role": "user",
        "content": "Now review the calling code that invokes process()."
      },
      {
        "role": "system",
        "content": "From now on, every suggestion must include explicit type annotations."
      }
    ]
  }'

Thinking mode is enabled by default

Passing in thinking: {"type": "adaptive"} is equivalent to not passing it at all, and the thinking content will be included in the response content in both cases.

Prompt cache threshold lowered

Minimum cache tokens: 512 (1024 for Opus 4.8), short prompts can be cached directly without modifying business code.

Disabling thinking requires effort to be highor lower

(The default value is high; if no effort parameter is passed, it also defaults to high.) On Claude Opus 5, only when the effort level is highor lower will the parameter thinking: {"type": "disabled"}be accepted. Setting xhighor maxeffort while including thinking: {"type": "disabled"}will result in a 400 error.This is a behavior that has been officially available starting from Claude Opus 5, and it is enforced on every request.