Unclear The finding does not say what triggers this control or who sets these values.
What
- The
thinking_displayparameter of theset_max_thinking_tokenstool now accepts a new mode, in addition tosummarizedandomitted, that shows one short title per stretch of thinking instead of a full prose summary. This mode only works for Claude Code sessions that Anthropic itself hosts; on any other client, the request still succeeds but silently falls back toomittedonce the API rejects the value. - A tightened validation schema now requires
max_thinking_tokensto be an integer ornull, andthinking_displayto be one of a fixed set of values ornull, with a matching error message when the input doesn't fit.
Why
The new display mode gives a more scannable view of Claude's thinking on supported sessions, while the stricter validation makes sure invalid values for these settings produce a clear error instead of being silently accepted.
Names in the bundlemax_thinking_tokensthinking_display
Documented inclaude-code/agent-sdk/python
The entry above is what we published on the day. These lines were added later, as Anthropic's own pages caught up, and they sit beside the original rather than replacing it.
Confirmed since
Anthropic's documentation has since written up max_thinking_tokens, on Agent SDK reference - Python.
| `max_thinking_tokens` | `int \| None` | `None` | *Deprecated* - Maximum tokens for thinking blocks. Use `thinking` instead |agent-sdk/python see the edit
Something disagreesSomething we can check disagrees with this entry, or the writer said they could not settle it.
The writer flagged doubt
The finding does not say what triggers this control or who sets these values.
Anthropic's documentation agrees
Anthropic's documentation has since written up max_thinking_tokens, on Agent SDK reference - Python.