Maximize Reasoning Depth with Always-On Adaptive Thinking
Enable Claude to 'think' longer for complex reasoning tasks.
Utilize Claude Fable 5's 'Adaptive Thinking' and 128k output tokens for complex software engineering and scientific research tasks that require sustained logic.
The Scenario
You are solving an intricate software engineering architectural problem or a complex scientific research query that requires deep, multi-step reasoning.
Before & after
Users had to manually prompt the AI to 'think step-by-step' or use chain-of-thought techniques to get accurate results for complex logic. This often required multiple prompt iterations over 15–20 minutes.
By selecting a model with 'Adaptive Thinking' (like Fable 5), the AI automatically allocates its 128k output tokens toward 'long-running cells' and deep reasoning. This setup takes seconds to initiate.
The Prompt
[USER_TASK_DESCRIPTION] Note: I am using Claude Fable 5. Please use your adaptive thinking capabilities to work through this multi-step logical problem, ensuring that you budget your internal reasoning before providing the final 128k-token-capable output.
Adaptive thinking is now 'always-on' in Fable 5, meaning the model naturally budgets its high output token limit (128k) for complex reasoning tasks without extra prompting.
Source
Claude Platform - Claude API Docs"Both models support a 1M token context window... and always-on adaptive thinking."