Python: feat: add thinking_level support for Gemini in Python

[markmcd]

Motivation and Context

Allow users to exercise control over how much they are willing to spend in terms of response latency and token budgets.

fixes: #13955

Description

Follows the same terminology and structure as used in the upstream API. Users provide an optional thinking_level param that is plumbed through to the genai SDK.

I have not added support for thinking_budget, which takes an integer, as it's not supported by newer models and will not be supported going forward. I have not updated the Vertex code path either as that's marked as deprecated in this library.

This matches the .NET implementation

Contribution Checklist

• The code builds clean without any errors or warnings
• The PR follows the SK Contribution Guidelines and the pre-submission formatting script raises no violations
• All unit tests pass, and I have added new tests where possible
• I didn't break anyone 😄

[Copilot]

Pull request overview

Adds support for configuring Gemini “thinking level” in the Python Google AI connector by introducing a thinking_level execution setting and mapping it to the GenAI SDK’s expected thinking_config request shape.

Changes:

• Added thinking_level to GoogleAIPromptExecutionSettings.
• Overrode prepare_settings_dict() to map thinking_level → thinking_config.thinking_level.
• Added unit tests covering the new field and the mapping behavior.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 3 comments.

File	Description
python/semantic_kernel/connectors/ai/google/google_ai/google_ai_prompt_execution_settings.py	Adds thinking_level setting and maps it into thinking_config for requests.
python/tests/unit/connectors/ai/google/google_ai/test_google_ai_request_settings.py	Adds unit tests for thinking_level storage and mapping into the request dict.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[markmcd]

Personally I prefer the | str option for forward-compatibility, while still enabling completions from the Literal, but happy to defer to maintainer style on this one.

[github-actions]

Automated Code Review

Reviewers: 4 | Confidence: 81%

✓ Correctness

The PR correctly adds thinking_level support for Google AI by defining the field on the base settings class and overriding prepare_settings_dict to remap it into the thinking_config structure expected by the Google GenAI SDK. The implementation is consistent with how other config fields (like tool_config) are handled in this codebase. No correctness issues found.

✓ Security Reliability

This PR adds a thinking_level field to Google AI settings and overides prepare_settings_dict to map it into the thinking_config structure expected by the Google SDK. The implementation is straightforward and low-risk. The value is passed as a structured parameter to the SDK (not interpolated into strings), so there's no injection risk. The | str in the type annotation disables Pydantic's Literal validation but appears intentional for forward-compatibility with future API levels. No resource leaks, unsafe deserialization, secrets, or unhandled failure modes are introduced.

✓ Test Coverage

The PR adds two unit tests that cover setting thinking_level and verifying the prepare_settings_dict mapping. While these cover the happy path, there are minor gaps: no test for the None/default case in prepare_settings_dict (ensuring thinking_config is absent), and no update to the existing default settings test to verify thinking_level defaults to None. The existing test_create_options implicitly exercises the None path through inheritance but doesn't explicitly assert thinking_config absence.

✓ Design Approach

I did not find a design-level issue that meets the evidence bar. The change introduces thinking_level on the Google AI prompt settings and remaps it into thinking_config at request-construction time, which is consistent with how the Google AI text/chat services consume prepare_settings_dict() before building GenerateContentConfigDict.

---

Automated review by markmcd's agents