LiteLLM DeepSeek thinking mode fails with missing `reasoning_conten` on tool calls

[Harrrryz]

Please read this first

• Have you read the docs? Agents SDK docs
• Have you searched for related issues? Yes

Describe the bug

When running the LiteLLM example against DeepSeek’s thinking mode with tools(DeepSeek V3.2 Thinking mode), the assistant tool-call message lacks the required reasoning_content field. DeepSeek responds with a 400 error: “Missing reasoning_content field in the assistant message …”.

Debug information

• Agents SDK version: current main (after latest changes)
• Python version: Python 3.13

Repro steps

uv run examples/model_providers/litellm_provider.py \
  --model deepseek/deepseek-reasoner \
  --api-key <deepseek_api_key>

The run fails after the tool call with a 400 complaining about missing reasoning_content.

Expected behavior

Assistant tool-call turns should include reasoning_content so DeepSeek can continue thinking + tool calling and return a final answer without error.