A new training recipe cuts hallucinations in multimodal reasoning models by trimming redundant thinking tokens and contrasting against deliberately induced errors.
document fetched 2026-10-28 02:52