fix(types): filter null fields from reasoning output items

[shivaaang]

Relevant issues

Fixes #16824

Pre-Submission checklist

• I have Added testing in the tests/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

Type

Bug Fix

Changes

Problem

When providers return reasoning output items without status, content, or encrypted_content, Pydantic's Optional defaults serialize them as null. This breaks downstream SDKs - for example, the OpenAI C# SDK crashes on status=null in reasoning items.

The request-side already handles this correctly via OpenAIResponsesAPIConfig._handle_reasoning_item(), which filters these fields before sending requests to providers. But the response-side had no equivalent filtering, so null fields leak into serialized responses.

Solution

Added a @field_serializer("output", mode="wrap") on ResponsesAPIResponse that removes null status, content, and encrypted_content from reasoning items during serialization. This mirrors the request-side filtering pattern.

• Only affects reasoning items (type == "reasoning"); all other output types (messages, function calls, etc.) are untouched
• Preserves non-null values (e.g., encrypted_content used by GitHub Copilot)
• Works for both model_dump() and model_dump_json() serialization paths

Files changed

• litellm/types/llms/openai.py - Added field_serializer import and _serialize_output_filter_reasoning_nulls method on ResponsesAPIResponse
• tests/test_litellm/types/llms/test_types_llms_openai.py - 7 new tests in TestResponsesAPIReasoningNullFields class

Testing

• 7 new tests covering null field removal (model_dump and model_dump_json), non-null value preservation, message items unaffected, mixed reasoning + message output, core field preservation, and top-level null fields unaffected
• All 16 tests in the file pass (9 existing + 7 new)
• Verified edge cases: empty output list, model_construct(), multiple reasoning items, exclude_none=True, mode="json"

@greptileai @shin-bot-litellm

[vercel]

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

Project	Deployment	Actions	Updated (UTC)
litellm	Ready	Preview, Comment	Feb 28, 2026 7:37am

[greptile-apps]

Greptile Summary

This PR fixes a downstream SDK compatibility issue (#16824) where null status, content, and encrypted_content fields on reasoning output items in ResponsesAPIResponse caused crashes in SDKs like the OpenAI C# SDK.

• Adds a @field_serializer("output", mode="wrap") on ResponsesAPIResponse that strips null values for status, content, and encrypted_content from reasoning items (type == "reasoning") during serialization
• The filtering logic is identical to the existing request-side filtering in OpenAIResponsesAPIConfig._handle_reasoning_item() (fallback path at litellm/llms/openai/responses/transformation.py:146-151), creating symmetry between request and response handling
• Non-reasoning output items (messages, function calls, etc.) and non-null values on reasoning items are unaffected
• Includes 7 focused unit tests with no network calls

Confidence Score: 5/5

• This PR is safe to merge — it's a targeted serialization fix with comprehensive test coverage and no side effects on non-reasoning output items.
• The change is minimal and well-scoped: a single field_serializer that only affects reasoning items during serialization. The filtering logic is battle-tested (identical to the existing request-side filtering). The 7 new tests cover all critical paths including model_dump, model_dump_json, mixed output, and preservation of non-null values. No existing behavior is altered for non-reasoning items.
• No files require special attention.

Important Files Changed

Filename	Overview
litellm/types/llms/openai.py	Adds a field_serializer on the output field of ResponsesAPIResponse to strip null status, content, and encrypted_content from reasoning items during serialization. The logic correctly mirrors the request-side filtering in _handle_reasoning_item().
tests/test_litellm/types/llms/test_types_llms_openai.py	Adds 7 well-structured unit tests covering null field removal, non-null preservation, message items unaffected, mixed output, core field preservation, and top-level fields. All tests are mock-only with no network calls.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A["ResponsesAPIResponse.model_dump() / model_dump_json()"] --> B["field_serializer('output', mode='wrap')"]
    B --> C["handler(value) — default serialization"]
    C --> D{"Is serialized output a list?"}
    D -- No --> E["Return as-is"]
    D -- Yes --> F["Iterate over items"]
    F --> G{"item is dict AND type == 'reasoning'?"}
    G -- No --> H["Keep item unchanged"]
    G -- Yes --> I["Filter out null status, content, encrypted_content"]
    I --> J["Return cleaned reasoning item"]
    H --> K["Return final output list"]
    J --> K

Loading

_{Last reviewed commit: 53350c8}

[greptile-apps]

_{2 files reviewed, no comments}

_{Edit Code Review Agent Settings | Greptile}