Why DeepSeek-R1 Ignores Thought-Token Budgets

Thoughtology research (arXiv:2504.07128) demonstrates that DeepSeek-R1 struggles to adhere to thought-token budgets specified inside the prompt text, and shows that trade-offs between budget compliance and accuracy exist. Do not ask R1 to cap its <think> word count in the prompt; let <think> run naturally and constrain only the final response format.

User Prompt: Analyze the thermodynamic efficiency limits of the combined-cycle turbine specification below. Present your final answer after </think> as a concise 4-row summary table.