models - litellm - start and stop reasoning

[pgrayy]

Description

We need to explicitly mark the start and stop of the reasoning content stream in the litellm model provider. With these changes, users can now do the following:

import json

from strands import Agent
from strands.models.litellm import LiteLLMModel

agent = Agent(
    model=LiteLLMModel(
        client_args={
            "api_key": "***",
        },
        model_id="anthropic/claude-sonnet-4-20250514",
        params={
            "thinking": {
                "budget_tokens": 1024,
                "type": "enabled",
            },
        },
    ),
)
result = agent("What is 2+2? Think through the answer step by step.")

print(json.dumps(result.message, indent=2))

The result of the above script will look something like:

{
  "role": "assistant",
  "content": [
    {
      "reasoningContent": {
        "reasoningText": {
          "text": "This is a very basic addition problem. Let me think through it step by step as requested.\n\n2 + 2\n\nI can think of this in several ways:\n\n1. Counting: If I have 2 objects, and I add 2 more objects, I can count them: 1, 2, 3, 4. So 2 + 2 = 4.\n\n2. Using the definition of addition: Addition means combining quantities. So if I have a quantity of 2, and I combine it with another quantity of 2, I get a total quantity of 4.\n\n3. Using basic arithmetic facts: This is one of the fundamental addition facts that 2 + 2 = 4.\n\nThe answer is 4."
        }
      }
    },
    {
      "text": "I'll work through this basic addition problem step by step:\n\n**Step 1: Understand what we're adding**\n- We have the number 2\n- We're adding another 2 to it\n- This can be written as: 2 + 2\n\n**Step 2: Think about what addition means**\n- Addition means combining or putting together quantities\n- We're combining a group of 2 with another group of 2\n\n**Step 3: Count or calculate**\n- We can count: 1, 2 (first group) + 1, 2 (second group) = 1, 2, 3, 4 total\n- Or we can recall that 2 + 2 is a basic math fact\n\n**Answer: 2 + 2 = 4**"
    }
  ]
}

Related Issues

#523

Documentation PR

N/A

Type of Change

Bug fix

Testing

How have you tested the change? Verify that the changes do not break functionality or introduce warnings in consuming repositories: agents-docs, agents-tools, agents-cli

• I ran hatch run prepare: Updated the unit tests.

Checklist

• I have read the CONTRIBUTING document
• I have added any necessary tests that prove my fix is effective or my feature works
• I have updated the documentation accordingly
• I have added an appropriate example to the documentation to outline the feature, or no new docs are needed
• My changes generate no new warnings
• Any dependent changes have been merged and published

---

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.

[zastrowm]

Would an integ test be helpful here? Ensuring that we get back reasoning content and normal content in the stream?

[Unshure]

Can a model switch between streaming reasoning content and text content in a single streamed response? If so, I think we might need to change this implementation:

Reasoning: Let me think about this problem
Text: I have thought about this problem, let me think about a solution
Reasoning: Im thinking about a solution
Text: I have thought about a solution

Not sure if any models follow this pattern?

[pgrayy]

Initially I tried to account for this but the code was a bit more complex. Since I wasn't aware of any model supporting this pattern, I decided instead to take a simpler approach and just implement for the pattern where reasoning is streamed first. I'm open to this being more flexible though.

[pgrayy]

Note, reasoning is supported in gemini but there is a similar bug to what we are fixing here for litellm. We need to yield a content stop block. Will address in a follow up PR.