test(vertex_ai): pin gemini-3.1-flash-lite reasoning mapping

[vincentkoc]

Relevant issues

Related: #22674, #22920

Pre-Submission checklist

Please complete all items before asking a LiteLLM maintainer to review your PR

• I have Added testing in the tests/test_litellm/ directory, Adding at least 1 test is a hard requirement - see details
• My PR passes all unit tests on make test-unit
• My PR's scope is as isolated as possible, it only solves 1 specific problem
• I have requested a Greptile review by commenting @greptileai and received a Confidence Score of at least 4/5 before requesting a maintainer review

CI (LiteLLM team)

• Branch creation CI run
  Link:

• CI run for the last commit
  Link:

• Merge / cherry-pick CI run
  Links:

Type

✅ Test

Changes

• Add explicit tests/test_litellm/ regression coverage for gemini-3.1-flash-lite-preview
• Cover all six supported reasoning_effort values for this model: minimal, low, medium, high, disable, none
• Assert the model stays on the thinkingLevel path and does not regress to thinkingBudget
• Assert includeThoughts stays aligned with the enabled vs disabled effort cases

Validation

• PYTEST_XDIST_WORKER=gw0 uv run pytest tests/test_litellm/llms/vertex_ai/gemini/test_vertex_and_google_ai_studio_gemini.py -k gemini_31_flash_lite -q
  • 6 passed, 87 deselected
• uv run ruff check tests/test_litellm/llms/vertex_ai/gemini/test_vertex_and_google_ai_studio_gemini.py
  • still blocked by pre-existing lint violations in this file unrelated to this diff

Reference

• https://docs.cloud.google.com/vertex-ai/generative-ai/docs/models/gemini/3-1-flash-lite

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
litellm	Error		Mar 8, 2026 5:48pm

[vincentkoc]

@Sameerlite this is a narrow follow-up to #22674 and #22920. It adds explicit tests/test_litellm/ coverage for gemini-3.1-flash-lite-preview reasoning mapping, based on the current Vertex AI model docs: https://docs.cloud.google.com/vertex-ai/generative-ai/docs/models/gemini/3-1-flash-lite

[greptile-apps]

Greptile Summary

This PR adds a single parametrized regression test (test_reasoning_effort_maps_to_thinking_level_gemini_31_flash_lite) that pins all six reasoning_effort → thinkingConfig mappings for gemini-3.1-flash-lite-preview in the tests/test_litellm/ directory.

Key points:

• All six reasoning_effort values (minimal, low, medium, high, disable, none) are covered with correct expected values verified against the implementation in _map_reasoning_effort_to_thinking_level.
• The test exercises the full map_openai_params integration layer, not just the private helper — consistent with the pattern used by the neighbouring test_reasoning_effort_maps_to_thinking_level_gemini_3 test.
• Three assertions per case: the exact thinkingLevel string, includeThoughts boolean, and the absence of thinkingBudget (confirming the model stays on the thinkingLevel code path).
• No real network calls; this is a pure-unit mock test as required by repository conventions.
• The PR description reports 6 passed, 87 deselected from a local run, satisfying the requirement for evidence that the issue is resolved.

Confidence Score: 5/5

• This is a test-only PR with no production code changes; it is safe to merge.
• The change is entirely additive test code. All expected parameter values were cross-checked against the _map_reasoning_effort_to_thinking_level implementation and are correct. The test uses the full integration path (map_openai_params) and makes no network calls. Previous review concerns (incomplete parametrize coverage, missing thinkingLevel assertion, bypassing the integration layer) are fully addressed in this revision.
• No files require special attention.

Important Files Changed

Filename	Overview
tests/test_litellm/llms/vertex_ai/gemini/test_vertex_and_google_ai_studio_gemini.py	Adds 6 parametrized regression tests for all reasoning_effort values for gemini-3.1-flash-lite-preview through the full map_openai_params integration layer; asserts thinkingLevel, includeThoughts, and absence of thinkingBudget.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A["test_reasoning_effort_maps_to_thinking_level_gemini_31_flash_lite\n(parametrized × 6 effort values)"] --> B["VertexGeminiConfig.map_openai_params(\n  model='gemini-3.1-flash-lite-preview'\n)"]
    B --> C{"_is_gemini_3_or_newer?"}
    C -- "Yes ('gemini-3' in model)" --> D["_map_reasoning_effort_to_thinking_level(effort, model)"]
    D --> E{"is_gemini3flash?\n('gemini-3.1-flash' in model)"}
    E -- "True" --> F["minimal → thinkingLevel='minimal'"]
    E -- "True" --> G["low → thinkingLevel='low'"]
    E -- "True" --> H["medium → thinkingLevel='medium'"]
    E -- "True" --> I["high → thinkingLevel='high'"]
    E -- "True" --> J["disable → thinkingLevel='minimal', includeThoughts=False"]
    E -- "True" --> K["none → thinkingLevel='minimal', includeThoughts=False"]
    F & G & H & I & J & K --> L["Assert thinkingConfig.thinkingLevel\nAssert thinkingConfig.includeThoughts\nAssert 'thinkingBudget' NOT in thinkingConfig"]

Loading

_{Last reviewed commit: 7d4bbcd}

[vincentkoc]

Addressed the Greptile follow-ups on the latest head:

• expanded the regression test to cover all six reasoning_effort values for gemini-3.1-flash-lite-preview
• verified the targeted test selection passes locally with:
  • PYTEST_XDIST_WORKER=gw0 uv run pytest tests/test_litellm/llms/vertex_ai/gemini/test_vertex_and_google_ai_studio_gemini.py -k gemini_31_flash_lite -q
  • result: 6 passed, 87 deselected

The PYTEST_XDIST_WORKER=gw0 env var avoids the repo-level module reload path that hits the known litellm.llms.base_llm.evals import issue in this environment.