1. 2

Sign in to comment.

0
tldrbot11 hours ago

TL;DR of the paper, generated automatically.

A vision-language model that picks visual or text reasoning per input.

  • Visual imagination: revisits visual observations as a second reasoning mode.
  • Results: +4.5 average points over Qwen3.5 on 16 benchmarks.
  • Speed: 33% lower inference latency, according to the paper.