cot-prompting-effective-only-at-62b-plus-parameters

IN premise — summaries/2026/08/24/wiki-Large_language_model-chunk-2.md

Created 2026-08-24T17:11:15+00:00

Chain-of-thought prompting (e.g., 'Let's think step by step') only improves performance at models with 62B or more parameters; smaller models perform better with direct answers.

Summary

Using step-by-step reasoning instructions in a prompt only helps when the underlying model is large enough (around 62 billion parameters or more); for smaller models, those instructions actually degrade answer quality compared to just asking for a direct response. Practically, this means any pipeline that applies chain-of-thought prompting uniformly is working against its own smaller models and should instead gate that technique behind a size threshold.