fix(thinking): handle None thinking param in is_thinking_enabled

[Terrajlz]

Summary

• Fix AttributeError: 'NoneType' object has no attribute 'get' when non_default_params["thinking"] is explicitly None
• Change non_default_params.get("thinking", {}).get("type") to (non_default_params.get("thinking") or {}).get("type") — handles both missing key AND explicit None
• Add unit tests covering: None, {"type": "enabled"}, missing key, empty dict, and {"type": "disabled"}

Fixes #28576

Test plan

• py -3.12 -m pytest tests/test_litellm/test_thinking_enabled.py -v — all 5 tests pass
• Verify with Anthropic-backed model via proxy with thinking: null in request

🤖 Generated with Claude Code + Cynthia

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you all sign our Contributor License Agreement before we can accept your contribution.
1 out of 3 committers have signed the CLA.

✅ yuneng-berri
❌ Cynthia Code
❌ shin-berri

---

Cynthia Code seems not to be a GitHub user. You need a GitHub account to be able to sign the CLA. If you have already a GitHub account, please add the email address used for this commit to your account.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[greptile-apps]

Greptile Summary

This PR fixes an AttributeError crash in is_thinking_enabled when the thinking key exists in non_default_params with an explicit None value — a case not handled by the previous .get("thinking", {}) pattern, which only substitutes the default when the key is absent.

• transformation.py: Replaces non_default_params.get("thinking", {}).get("type") with (non_default_params.get("thinking") or {}).get("type"), correctly short-circuiting any falsy value (None, False, 0, "") to an empty dict before calling .get("type").
• test_thinking_enabled.py: Adds 9 parameterized unit tests covering None, {"type": "enabled"}, missing key, empty dict, {"type": "disabled"}, reasoning_effort, combined params, and other falsy values — all exercised without real network calls.

Confidence Score: 5/5

Safe to merge — the change is a minimal, targeted guard for a well-understood crash path with no behavioral changes for existing valid inputs.

The one-line fix correctly handles the None case without touching any other logic, and the new tests cover the full range of falsy inputs as well as the happy path. No existing tests were modified, no network calls are made in the test suite, and the change is confined to a single non-critical utility method.

No files require special attention.

Important Files Changed

Filename	Overview
litellm/llms/base_llm/chat/transformation.py	One-line fix: replaces .get("thinking", {}) with (... or {}) so an explicit None value no longer causes AttributeError
tests/test_litellm/test_thinking_enabled.py	New unit test file covering 9 parameterized cases (None, enabled, missing, empty dict, disabled, reasoning_effort, combined, and falsy values); no real network calls

_{Reviews (1): Last reviewed commit: "fix(thinking): handle None thinking para..." | Re-trigger Greptile}

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[oss-pr-review-agent-shin]

🤖 litellm-agent: Squash-merged into staging branch shin_agent_oss_staging_05_22_2026. Staging PR: #28601

---

Triage Summary
Guards is_thinking_enabled in BaseConfig against an AttributeError crash when non_default_params["thinking"] is explicitly None. Replaces .get("thinking", {}) with (... or {}) so any falsy value short-circuits to an empty dict before .get("type") is called. Adds a new test file with 9 parameterized cases covering None, enabled, missing key, empty dict, disabled, reasoning_effort, and other falsy values.

Merge Confidence: 5/5 ✅ READY
⚠️ 1 check failing: lint
Ready to ship.

Greptile 5/5, no blocking pattern findings, no CircleCI runs (OSS-typical). 1 check failing but unrelated to this diff: lint. 1 check also red on neighboring PRs (lint) — infra-wide noise, no penalty.