fix(gemini): use longest-prefix match for context window lookup

[a-yeyang]

Summary

• GeminiCompletion.get_context_window_size() returned the first matching prefix in its lookup table via startswith, so the more specific "gemini-2.0-flash-thinking" entry was always shadowed by the shorter "gemini-2.0-flash" entry listed above it — that row was dead code.
• Replaced the first-match loop with a longest-matching-prefix selection, so entry order in the table no longer matters and a more specific prefix always wins over a shorter one.

Validation

• Added test_gemini_context_window_size_prefers_longest_matching_prefix in lib/crewai/tests/llms/google/test_google.py, asserting gemini-2.0-flash-thinking-exp-01-21 resolves to the 32768-token window rather than the 1048576-token gemini-2.0-flash window.
• Ran the full lib/crewai/tests/llms/google/ suite: 52 passed, 2 skipped (1 pre-existing unrelated failure due to a missing litellm extra in the local dev env, reproduced identically on main).
• ruff check / ruff format --check pass on both changed files.

Fixes #7129

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 431456b8-0f62-427f-80bf-3007b0248e69

📥 Commits

Reviewing files that changed from the base of the PR and between fcdeb3d and b9450b1.

📒 Files selected for processing (2)

• lib/crewai/src/crewai/llms/providers/gemini/completion.py
• lib/crewai/tests/llms/google/test_google.py

Included review availability: Your plan provides up to 10 included reviews per hour; 9 remain after this review.

---

📝 Walkthrough

Walkthrough

The Gemini context window lookup now selects the longest matching model prefix. A regression test verifies the specific thinking model value and the standard flash model value.

Changes

Gemini context window resolution

Layer / File(s)	Summary
Longest-prefix lookup and regression coverage lib/crewai/src/crewai/llms/providers/gemini/completion.py, lib/crewai/tests/llms/google/test_google.py	The lookup selects the longest matching model prefix. Tests verify the expected context windows for thinking and standard flash models.

Suggested reviewers: lucasgomide, lorenzejay

Merge Risk: ⚪ Minimal · up to b9450

The change makes Gemini select the most specific configured context window for a model prefix and adds regression coverage; no actionable merge-blocking risk remains beyond normal checks and review.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly and concisely describes the main change: Gemini context window lookup now uses the longest matching prefix.
Description check	✅ Passed	The description explains the Gemini prefix-matching bug, the implemented fix, the regression test, and validation results. It is directly related to the changeset.
Linked Issues check	✅ Passed	The changes satisfy issue #7129. Gemini models using the "gemini-2.0-flash-thinking" prefix now select the specific entry and return int(32768 * 0.85), while the regression test verifies the behavior.
Out of Scope Changes check	✅ Passed	The changes are limited to the Gemini context-window lookup and its regression test. No unrelated code changes are identified.
Docstring Coverage	✅ Passed	Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 4 functions across 2 files.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}