Preprint claims latent diffusion unifies multimodal reasoning, unreplicated

The Uni-LaDiR preprint self-reports a 7.3% relative gain over the strongest evaluated baselines across 11 vision-language benchmarks, and 6.1% across 2 robot manipulation suites.

Method-wise, teacher reasoning steps are mapped by a unified encoder into shared latent-space tokens, and a diffusion model predicts the next block of thought tokens; at inference no teacher observations are used. This differs from the common approach of concatenating modality-specific tokens.

Note that all numbers are first-party evaluations with author-selected baselines and no independent replication; the paper was submitted September 17 and revised to v2 on September 23.