Anthropic publishes Reasoning models don't always say what they think
Published Reasoning models don't always say what they think — the rate of mentioning the hint was 25% for Claude 3.7 Sonnet and 39% for DeepSeek R1
Why it matters
Checked, using its own model, that CoT unfaithfulness persists even in the era of reasoning models
Source — Primary sources first; secondary reports are labelled “Press”
Officialanthropic.comhttps://www.anthropic.com/research/reasoning-models-dont-say-thinkAdded in Week 35, 2026 · See that week