litellm - 💡(How to fix) Fix use_chat_completions_api: true drops content when provider returns reasoning_content [1 pull requests]

Bug Description

When use_chat_completions_api: true is configured on a model with model: openai/..., the proxy correctly calls the upstream OpenAI API's /v1/chat/completions endpoint. However, when converting the response back to Anthropic Messages API format, the content field is dropped when the upstream also returns reasoning_content (a non-standard field used by DeepSeek, Kimi, etc. for chain-of-thought).

Expected

The Anthropic-format response should include the model text content:

{"type":"message","content":[{"type":"text","text":"hello"}],"stop_reason":"end_turn"}

Actual

Response has empty content when upstream includes reasoning:

{"type":"message","content":[],"stop_reason":"end_turn"}

Steps

1. Config:

model_list:
  - model_name: deepseek-v4-flash
    litellm_params:
      model: openai/deepseek-v4-flash
      api_base: https://opencode.ai/zen/go/v1
      api_key: os.environ/OPENCODE_GO_API_KEY
      use_chat_completions_api: true

1. Start proxy: litellm --config config.yaml --port 4000

2. Send request via /v1/messages (Anthropic): curl http://localhost:4000/v1/messages -H "Authorization: Bearer sk-test" -H "anthropic-version: 2023-06-01" -d '{"model":"deepseek-v4-flash","max_tokens":200,"messages":[{"role":"user","content":"say hello"}]}'

3. Result: empty content

4. Same request via /v1/chat/completions (OpenAI) works and returns content: "hello" with reasoning_content: "Thinking...".

The reasoning_content field triggers a code path that leaves the final message with empty content.

Environment

LiteLLM 1.83.14, Python 3.13.3, macOS Affected providers: DeepSeek (opencode.ai), Kimi (CanopyWave) - any provider returning reasoning_content