fix(anthropic-adapter): correct streaming for reasoning_content and empty tool_calls

[lukesorvikDO]

Testing Doc : https://docs.google.com/document/d/1LpUJYUc2a2s5iMxbg_6JMiqQEc0sDDOuTeJvVqonMB0/edit?tab=t.f2qn8vnp2dx9#heading=h.xf1r939ub9pb

Original pr: #21

Relevant issues

Fixes invalid Anthropic SSE from the experimental pass-through adapter when models return reasoning_content. Claude Code rejects the stream (Content block is not a thinking block) and retries with a non-streaming request, causing 2× POST /v1/messages per turn through inference-proxy.

Problem
When a model returns reasoning_content (e.g. reasoning models), the adapter opened a text content block but emitted thinking_delta events. That violates the Anthropic SSE contract. Claude Code aborts the stream and falls back to non-streaming → duplicate requests and double routing cost.

Local live test
Setup: pip install -e ".[proxy]" in venv, Ollama deepseek-r1:1.5b at http://localhost:11434

python - <<'PY'
import asyncio, litellm
async def main():
    stream = await litellm.anthropic_messages(
        model="ollama/deepseek-r1:1.5b",
        api_base="http://localhost:11434",
        max_tokens=128, stream=True,
        messages=[{"role": "user", "content": "Say hi in one short sentence."}],
    )
    async for chunk in stream:
        line = chunk.decode() if isinstance(chunk, bytes) else str(chunk)
        if "content_block_start" in line or "thinking_delta" in line:
            print(line[:200])
asyncio.run(main())
PY

Before fix ❌

event: content_block_start
data: {"type": "content_block_start", "index": 0, "content_block": {"type": "text", "text": ""}}

event: content_block_delta
data: {"type": "content_block_delta", "index": 0, "delta": {"type": "thinking_delta", "thinking": "Alright"}}

event: content_block_delta
data: {"type": "content_block_delta", "index": 0, "delta": {"type": "thinking_delta", "thinking": ","}}

event: content_block_delta
data: {"type": "content_block_delta", "index": 0, "delta": {"type": "thinking_delta", "thinking": " the"}}
...

thinking_delta on a text block → invalid SSE.

After fix ✅

event: content_block_start
data: {"type": "content_block_start", "index": 0, "content_block": {"type": "thinking", "thinking": ""}}

event: content_block_delta
data: {"type": "content_block_delta", "index": 0, "delta": {"type": "thinking_delta", "thinking": "Alright"}}

event: content_block_delta
data: {"type": "content_block_delta", "index": 0, "delta": {"type": "thinking_delta", "thinking": ","}}

event: content_block_delta
data: {"type": "content_block_delta", "index": 0, "delta": {"type": "thinking_delta", "thinking": " so"}}
...

content_block_start correctly opens a thinking block before thinking_delta events.

Unit tests

python -m pytest tests/test_litellm/llms/anthropic/experimental_pass_through/ -q
# 197 passed (on release/v1.83.10-stable.do.0)

Linear ticket

Pre-Submission checklist

Please complete all items before asking a LiteLLM maintainer to review your PR

• I have added meaningful tests
• 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

Delays in PR merge?

If you're seeing a delay in your PR being merged, ping the LiteLLM Team on Slack (#pr-review).

CI (LiteLLM team)

CI status guideline:

• 50-55 passing tests: main is stable with minor issues.
• 45-49 passing tests: acceptable but needs attention
• <= 40 passing tests: unstable; be careful with your merges and assess the risk.

• Branch creation CI run
  Link:

• CI run for the last commit
  Link:

• Merge / cherry-pick CI run
  Links:

Screenshots / Proof of Fix

Type

🆕 New Feature
🐛 Bug Fix
🧹 Refactoring
📖 Documentation
🚄 Infrastructure
✅ Test

Changes

[ShivaniKumar1]

Empty reasoning_content can open a thinking block without subsequent thinking_delta, violating Anthropic streaming sequence and breaking clients. Can you make sure this is covered?