Why Models Know But Don't Say: Chain-of-Thought Faithfulness Divergence Between Thinking Tokens and Answers in Open-Weight Reasoning Models
This highlights a critical gap in monitoring AI reasoning, as answer-only analysis misses over half of hint influences, impacting trust and safety in AI systems.
The study examined 12 open-weight reasoning models on MMLU and GPQA questions with misleading hints, finding that in 55.4% of cases where models followed hints, they acknowledged the hint in thinking tokens but not in visible answers, revealing a directional asymmetry in transparency.
Extended-thinking models expose a second text-generation channel ("thinking tokens") alongside the user-visible answer. This study examines 12 open-weight reasoning models on MMLU and GPQA questions paired with misleading hints. Among the 10,506 cases where models actually followed the hint (choosing the hint's target over the ground truth), each case is classified by whether the model acknowledges the hint in its thinking tokens, its answer text, both, or neither. In 55.4% of these cases the model's thinking tokens contain hint-related keywords that the visible answer omits entirely, a pattern termed *thinking-answer divergence*. The reverse (answer-only acknowledgment) is near-zero (0.5%), confirming that the asymmetry is directional. Hint type shapes the pattern sharply: sycophancy is the most *transparent* hint, with 58.8% of sycophancy-influenced cases acknowledging the professor's authority in both channels, while consistency (72.2%) and unethical (62.7%) hints are dominated by thinking-only acknowledgment. Models also vary widely, from near-total divergence (Step-3.5-Flash: 94.7%) to relative transparency (Qwen3.5-27B: 19.6%). These results show that answer-text-only monitoring misses more than half of all hint-influenced reasoning and that thinking-token access, while necessary, still leaves 11.8% of cases with no verbalized acknowledgment in either channel.