Anthropic Claude 4.6 Sonnet Documentation

Claude 4.6 Sonnet is engineered for rapid, iterative development and stateful agentic workflows. It introduces persistent memory mechanisms for end-to-end project management and delivers high-reliability computer use capabilities, making it ideal for continuous integration, browser automation, and complex backend refactoring.

Key Capabilities

Iterative Development: Perfect for rapid feedback loops, code modifications, and continuous integration pipelines.
Complex Codebase Navigation: Seamlessly maps dependencies and tracks state changes across multi-directory projects.
Project Management with Memory: Retains contextual project memory across extended sessions to track historical engineering decisions.
Polished Document Creation: Generates production-ready technical specifications, system design schemas, and professional documentation.
Advanced Computer Use: Autonomously interacts with web interfaces, executing QA testing, scraping, and end-to-end workflow automation.

Request Parameters

To interact with this model via an OpenAI-compatible gateway or OpenRouter endpoint:

Parameter	Type	Required	Description
`model`	`string`	Yes	Use `"claude-sonnet-4-6"` or `"anthropic/claude-4.6-sonnet"`.
`messages`	`array`	Yes	Array of standard conversation objects (role/content).
`max_tokens`	`integer`	No	Limits the generation length. Default: `4096`.
`temperature`	`float`	No	Recommended: `0.2` for coding; `0.7` for document polishing.

Thinking Mode Configuration

claude-sonnet-4-6 supports Adaptive Thinking — the model dynamically decides when and how much to think based on task complexity. The legacy "type": "enabled" + budget_tokens parameters are deprecated for this model.

Important Constraints

If thinking is included, temperature must be 1.0 (or omitted — it defaults to 1.0).
max_tokens must be large enough to accommodate both the internal reasoning and the visible response (e.g., 16000).

Effort Levels

Value	Description
`"low"`	Fast responses; minimal thinking for simple sub-tasks
`"medium"`	Balanced speed and reasoning depth
`"high"`	Deep reasoning for complex tasks (Default)
`"max"`	Maximum reasoning capacity for the hardest problems

Request Example

json Copy

{
  "model": "claude-sonnet-4-6",
  "max_tokens": 16000,
  "thinking": {
    "type": "adaptive"
  },
  "output_config": {
    "effort": "high"
  },
  "messages": [
    {
      "role": "user",
      "content": "Your question here"
    }
  ]
}

Note: Deprecated format — do NOT use "type": "enabled" or budget_tokens with this model. Use "type": "adaptive" combined with output_config.effort instead.

claude-sonnet-4-6

Input

Output