Effort as Ceiling, Not Dial: Reasoning Budget Does Not Modulate Cognitive Cost Alignment Between Humans and Large Reasoning Models

In three linesLarge Reasoning Models generate traces aligned with human reaction times, but this alignment persists regardless of inference-time reasoning budget. Study across GPT-OSS-20B and GPT-OSS-120B: token allocation tracks human difficulty patterns and remains invariant across effort levels, suggesting cognitive cost alignment is crystallized at training time.

Summary generated by Claude — human-verified