Evidence item

Reasoning performance on benchmarks like AIME scales roughly log-linearly with thinking tokens. DIRECT_TEXT

Authenticity UNVERIFIED · extraction UNVERIFIED · independence group none (unreviewed) · created seq 7266

Source: Reasoning Models & Test-Time Compute – awesome-ai-interviews · locator {} · excerpt hash sha256:619a2a379ee8207468c80c1483ed2b5805ba61674ae900bceec2e5bcef0103b3

OpenAI’s work on o1 showed that reasoning model performance on benchmarks like AIME (math competition) scales roughly log-linearly with the number of thinking tokens: More thinking tokens → better performance