Model scale and certainty do not predict multi-agent persuasion dynamics
October 5, 2026
An evaluation of seven open-weight models shows that receivers frequently abandon initial judgments after a single exchange with a dissenting peer. Findings indicate that model scale and standalone certainty are poor predictors of whether a model will be persuaded or act as a persuader.
HOW THIS AFFECTS YOU
●
builderBe aware that small models can be just as influential or resistant as larger models in multi-agent workflows.
●
researcherYou cannot rely on model size to predict stability in multi-agent interaction studies.