Claude Models Reasoning and Web Search

Knox has backend adapters for the Claude model IDs anthropic/claude-sonnet-4.6, anthropic/claude-opus-4.6, and anthropic/claude-opus-4.7. These adapters can translate Knox request fields into Anthropic adaptive thinking settings and can also inject the Anthropic web-search tool when the request is routed through an AI gateway-backed channel.

Use /v1/chat/completions if you want reasoning content and web citations in the response. /v1/completions is available for compatibility, but it returns plain text only.

Supported adaptive reasoning levels

Model	Allowed reasoning_effort values	Backend default
anthropic/claude-sonnet-4.6	low, medium, high	high
anthropic/claude-opus-4.6	medium, high, max	high
anthropic/claude-opus-4.7	high, xhigh, max	xhigh

When you send reasoning_effort, Knox translates it into an Anthropic thinking block like this:

{
  "thinking": {
    "type": "adaptive",
    "effort": "high"
  }
}

Web search behavior

For Anthropic models routed through Knox channels:

• web_search defaults to true on both /v1/chat/completions and /v1/completions.
• Set "web_search": false to disable live search for a request.
• In chat completions responses, search citations are normalized into message.annotations using OpenAI-style url_citation entries.

Chat Completions Example

This is the best choice when you want both reasoning output and search citations back from the API.

cURL

curl https://api.knox.chat/v1/chat/completions \
  -H "Authorization: Bearer $KNOXCHAT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "anthropic/claude-sonnet-4.6",
    "messages": [
      {
        "role": "user",
        "content": "Summarize the latest changes in the EU AI Act and cite sources."
      }
    ],
    "reasoning_effort": "high",
    "web_search": true,
    "max_tokens": 1200
  }'

Python

from openai import OpenAI

client = OpenAI(
    base_url="https://api.knox.chat/v1",
    api_key="<KNOXCHAT_API_KEY>",
)

response = client.chat.completions.create(
    model="anthropic/claude-sonnet-4.6",
    messages=[
        {
            "role": "user",
            "content": "Summarize the latest changes in the EU AI Act and cite sources.",
        }
    ],
    reasoning_effort="high",
    web_search=True,
    max_tokens=1200,
)

message = response.choices[0].message
print(message.reasoning)
print(message.annotations)
print(message.content)

TypeScript

import OpenAI from 'openai';

const client = new OpenAI({
  baseURL: 'https://api.knox.chat/v1',
  apiKey: '<KNOXCHAT_API_KEY>',
});

const response = await client.chat.completions.create({
  model: 'anthropic/claude-sonnet-4.6',
  messages: [
    {
      role: 'user',
      content: 'Summarize the latest changes in the EU AI Act and cite sources.',
    },
  ],
  reasoning_effort: 'high',
  web_search: true,
  max_tokens: 1200,
});

const message = response.choices[0].message;
console.log(message.reasoning);
console.log(message.annotations);
console.log(message.content);

Response shape with citations

{
  "choices": [
    {
      "message": {
        "role": "assistant",
        "reasoning": "Search current legislative sources, reconcile drafts versus published text, then summarize by compliance impact.",
        "content": "The latest updates focus on implementation timing, GPAI obligations, and enforcement milestones.",
        "annotations": [
          {
            "type": "url_citation",
            "url_citation": {
              "url": "https://example.com/source",
              "title": "Source title"
            }
          }
        ]
      }
    }
  ]
}

Opus example

Use a higher reasoning level for the larger Opus models:

{
  "model": "anthropic/claude-opus-4.7",
  "messages": [
    {
      "role": "user",
      "content": "Research the current state of battery supply chains and produce a risk briefing with sources."
    }
  ],
  "reasoning_effort": "xhigh",
  "web_search": true,
  "max_tokens": 2000
}

Advanced passthrough: explicit thinking

If you want to control the Anthropic-native payload directly, Knox also accepts a thinking object and forwards it upstream:

{
  "model": "anthropic/claude-opus-4.6",
  "messages": [
    {
      "role": "user",
      "content": "Evaluate the tradeoffs of building an internal search index versus using managed search."
    }
  ],
  "thinking": {
    "type": "adaptive",
    "effort": "max"
  },
  "web_search": false
}

Use this only when you want raw Anthropic-style control. For most Knox clients, reasoning_effort is the simpler interface.

Using /v1/completions

Knox also supports legacy prompt-based access for these Claude models:

curl https://api.knox.chat/v1/completions \
  -H "Authorization: Bearer $KNOXCHAT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "anthropic/claude-opus-4.6",
    "prompt": "Find the latest SOC 2 guidance updates and summarize the implementation impact.",
    "reasoning_effort": "max",
    "web_search": true,
    "max_tokens": 900
  }'

On this route, Knox converts the prompt into chat messages internally and may still perform search and adaptive thinking upstream. The returned payload is converted into text completion format, so choices[].text is preserved, but message.reasoning and message.annotations are not.

Practical guidance

• Use /v1/chat/completions when you need reasoning output or citations.
• Set web_search: false if you want Claude reasoning without live search.
• Prefer reasoning_effort for normal Knox usage, and use thinking only for direct Anthropic-style control.
• anthropic/claude-opus-4.7 supports the highest default reasoning level in the current backend mapping.