[FEATURE] Improve handling for models not supporting reasoning content

[Luke-Shepp]

Problem Statement

In a flow where an initial agent acts as a router and classifies the users message and hands off to the appropriate model there are problems passing through message context / reasoning content.

For example, a model as a prompt router that can route between two agents using openai.gpt-oss-20b-1:0 and global.anthropic.claude-haiku-4-5-20251001-v1:0, depending on user message. This flow currently fails occasionally with:

An error occurred (ValidationException) when calling the ConverseStream operation: The model returned the following errors: messages.1.content.0.thinking.signature: Field required

A search for the error reveals this comment stating that Anthropic models can only process reasoning from Anthropic models.

Within the Strands SDK there is already a block to remove reasoning content for Deepseek models:

sdk-python/src/strands/models/bedrock.py

Lines 318 to 322 in 7cd10b9

	# DeepSeek models have issues with reasoningContent
	# TODO: Replace with systematic model configuration registry (https://github.com/strands-agents/sdk-python/issues/780)
	if "deepseek" in self.config["model_id"].lower() and "reasoningContent" in content_block:
	dropped_deepseek_reasoning_content = True
	continue

Proposed Solution

No response

Use Case

This should be configurable per model, even if it is something as simple as;

haiku_45 = BedrockModel(
    model_id="global.anthropic.claude-haiku-4-5-20251001-v1:0",
    region_name=os.environ.get("AWS_REGION"),
    temperature=0.0,
    max_tokens=2048,
+   drop_reasoning=True
)

Alternatives Solutions

No response

Additional Context

No response

[Luke-Shepp]

To add to this;

https://docs.aws.amazon.com/bedrock/latest/userguide/claude-messages-thinking-encryption.html

Claude doesn't require thinking blocks at all unless you're using extended thinking.