[FEATURE] Support Claude 4.6 Adaptive Thinking and Effort Parameter

[statefb]

Problem Statement

Claude 4.6 introduces new thinking features that deprecate the current budget_tokens approach:

Deprecated (will be removed in future models):

• thinking: {type: "enabled", budget_tokens: N}

New recommended approach:

• thinking: {type: "adaptive"} - Claude dynamically decides when and how much to think
• effort parameter (low, medium, high, max) - Controls thinking depth

Currently, Strands SDK only supports the deprecated budget_tokens via additional_request_fields, with no native support for the new
Claude 4.6 features.

Reference: https://platform.claude.com/docs/en/about-claude/models/whats-new-claude-4-6

Proposed Solution

No response

Use Case

1. Agentic Applications with Dynamic Reasoning

• Agents that need to think deeply on complex tasks but skip thinking for simple ones
• thinking: {type: "adaptive"} with effort: "high" provides optimal cost/quality balance

2. Cost Optimization

• effort: "low" for simple queries (faster, cheaper)
• effort: "max" for complex reasoning tasks (best quality)

3. Future-Proofing

• budget_tokens will be removed in future Claude models
• Applications using Strands need a migration path to the new API

Example usage:

from strands import Agent                                                                                                                  
from strands.models import BedrockModel
                                                                                                                                           
model = BedrockModel(
    model_id="anthropic.claude-opus-4-6-20250514-v1:0",                                                                                    
    additional_request_fields={                                                                                                            
        "thinking": {"type": "adaptive"},
        "effort": "high"
    }
)

agent = Agent(model=model, tools=[...])
result = agent("Solve this complex problem...")

Alternatives Solutions

No response

Additional Context

No response