Claude Code Thinking Mode: What It Means and Which Option to Use

“Claude Code thinking mode” can refer to different things: asking Claude Code to plan before editing, adjusting its effort, or viewing thinking-related output. Those are not the same as configuring a model request through the Messages API. Choose based on your goal: a plan, more interface visibility, or thinking in your own application.

If you want to…	Use…	Keep in mind…
Compare approaches before code changes	Plan Mode and a specific prompt	A plan helps you review options; it does not guarantee a correct implementation.
Adjust the model or its effort in Claude Code	/model to choose a model; /effort to set effort	They are separate controls. Effort does not select the model.
See thinking-related output in Claude Code	A summary setting, verbose output, or transcript view	These are display paths, not API settings or a guarantee of a complete reasoning record.
Use thinking in an application	A model-supported Messages API configuration	Check compatibility and account for thinking within output-token limits and usage.

In Claude Code: plan, adjust effort, or view output

When a change has several plausible designs, ask Claude Code to compare them before editing. For example:

Compare the two smallest ways to add account-level rate limits. For each, list the files it would change, likely failure modes, and a command to verify the result. Do not edit files; wait for approval.

This asks for decision-relevant alternatives and verification rather than a vague “think harder” instruction. Review the plan, then verify the implementation; a plan is not proof that a change is correct. Anthropic’s Claude Code command guide describes /plan as switching into plan mode. Use /model to select a model and /effort to adjust effort. The opusplan alias is documented as using Opus in plan mode and Sonnet for execution; verify the model resolution for your provider and setup before relying on that alias.

Viewing thinking-related output

For a terminal interactive session, a Claude Code collaborator described three display paths in an August 17, 2026 comment:

1. Inline summaries: add "showThinkingSummaries": true to your user settings file, ~/.claude/settings.json. Merge the key into your existing JSON rather than replacing the file:

   {
     "showThinkingSummaries": true
   }
   

2. Fuller blocks in the main view: enable Verbose output in /config, or add "verbose": true to settings. One user reported unrelated log noise with verbose output in Claude Code v2.0.36, so it may be less focused than summaries.

3. Transcript inspection: the collaborator said Ctrl+O opens transcript mode and displays thinking as it is generated. Keyboard shortcuts can vary by platform and terminal; consult the shortcut help in your interface if Ctrl+O behaves differently.

These are instructions from a collaborator comment, not a verified version-by-version settings reference. Check the current settings documentation for your installed version. Displaying a summary or block does not itself enable a model capability or guarantee access to a complete reasoning trace.

In a production API: use model-supported thinking

For a direct Anthropic API integration, the current Claude Sonnet 5.5 migration guide provides a documented Messages SDK example using model claude-sonnet-5-5 and output_config effort set to medium. Sonnet 5.5 uses adaptive thinking by default; the guide also accepts an explicit thinking: {"type": "adaptive"} configuration. This example uses the official Python SDK, which handles request headers:

from anthropic import Anthropic

client = Anthropic()
response = client.messages.create(
    model="claude-sonnet-5-5",
    max_tokens=4096,
    thinking={"type": "adaptive"},
    output_config={"effort": "medium"},
    messages=[
        {
            "role": "user",
            "content": "Find the likely cause of this failing test and suggest a verification step.",
        }
    ],
)

for block in response.content:
    if block.type == "text":
        print(block.text)

For raw HTTP requests, the Messages API reference labels 2026-10-21 as its current API version and lists the anthropic-version header as optional. Do not treat that date as a required header for the SDK example; consult API versioning guidance when pinning a version in a raw request.

Handle the response by content-block type rather than assuming the first block is text: a response can begin with a thinking block, and the model may omit one when it does not think a request needs it. Thinking tokens count toward max_tokens and are billed as output tokens, even when the thinking text is not returned. Leave output capacity for both reasoning and the final answer, and monitor actual usage.

Compatibility matters. On Sonnet 5.5, adaptive thinking is on by default and effort controls thinking depth; the model guide lists high as the API default. Older fixed-budget configurations are not interchangeable across models: Anthropic says fixed-budget thinking is deprecated on Claude 4.6 and rejected by Claude 4.7 and later. Check the exact model’s current guide before copying settings from older examples.

Which option should you choose?

• You need to compare code designs: use /plan, ask for alternatives and verification steps, then review before editing.
• You want to change the model or effort in Claude Code: use /model for the model and /effort for effort.
• You want to see thinking-related output in an interactive session: use the summary setting, verbose output, or transcript view described above, subject to your version and interface.
• You are building an application: configure a supported thinking mode and effort in the Messages API, confirm compatibility for the exact model, and budget for output tokens.

In short: planning shapes the workflow, display settings shape what the interface shows, and API parameters configure a supported model request. They are not interchangeable, and none guarantees correctness.

Sources

Claude Sonnet 5.5 migration guide — Claude Platform Docs

What's new in Claude Sonnet 5.5 — Claude Platform Docs

Claude Sonnet 5.5 — Claude Platform Docs

Create a Message — Claude API Reference

Adaptive thinking — Claude Platform Docs

Thinking — Claude Platform Docs

Troubleshooting thinking — Claude Platform Docs

Commands — Claude Code Docs

Model configuration — Claude Code Docs

Settings files and precedence — Claude Code Docs

All settings — Claude Code Docs

Interactive mode — Claude Code Docs

[FEATURE] Add Option to Always Show Claude's Thinking — GitHub