[DOCS] Thinking docs omit `--thinking disabled` and per-model disable behavior for default-thinking models

Documentation Type

Unclear/confusing documentation

Documentation Location

https://code.claude.com/docs/en/model-config

Section/Topic

Extended thinking controls and disabled-thinking behavior

Current Documentation

The model configuration page currently lists three controls:

"Toggle for the current session"

"Set the global default"

"Disable regardless of effort"

For disabling, it only says:

"Set MAX_THINKING_TOKENS=0."

The environment variables page also says:

"MAX_THINKING_TOKENS ... Set to 0 to disable thinking entirely."

What's Wrong or Missing?

Claude Code v2.1.166 clarifies or changes thinking disable behavior:

• MAX_THINKING_TOKENS=0 disables thinking on models that think by default via the Claude API.
• --thinking disabled also disables thinking.
• The per-model thinking toggle also disables thinking.
• Third-party providers are unchanged.

The current model configuration docs mention MAX_THINKING_TOKENS=0, but they do not document --thinking disabled, do not name the per-model thinking toggle, and do not explain the "models that think by default via the Claude API" scope or the third-party-provider exception.

As written, users cannot tell which disable mechanism to use for a one-session CLI launch, whether the per-model toggle is equivalent, or whether the behavior is provider-specific.

Suggested Improvement

Update the "Extended thinking" controls table to include the missing controls:

| Control | How to set it |
|---|---|
| Disable for one launch | Start Claude Code with --thinking disabled |
| Disable with environment | Set MAX_THINKING_TOKENS=0 |
| Disable per model | Use the per-model thinking toggle in the model picker |

Add a short scope note:

On models that think by default via the Claude API, these controls disable thinking. Third-party provider behavior is unchanged.

Also add or update the CLI reference row for --thinking if that flag is intended to be public.

Impact

Medium - Makes feature difficult to understand

Additional Context

Affected Pages:

| Page | Line(s) | Context |
|------|---------|---------|
| https://code.claude.com/docs/en/model-config | 239-247 | Extended thinking controls omit --thinking disabled and per-model disable semantics |
| https://code.claude.com/docs/en/env-vars | 309 | MAX_THINKING_TOKENS=0 is documented but not scoped to the new default-thinking behavior |
| https://code.claude.com/docs/en/cli-reference | 47-149 | CLI flags table does not show --thinking disabled |
| https://code.claude.com/docs/en/desktop | 523 | Desktop docs mention only MAX_THINKING_TOKENS=0 |

Total scope: 4 pages affected.

Version reference: Claude Code v2.1.166 release notes state that MAX_THINKING_TOKENS=0, --thinking disabled, and the per-model thinking toggle now disable thinking on models that think by default via the Claude API, with third-party providers unchanged.