arXivZiyu Guo, Rain Liu, Xinyan Chen, Pheng-Ann HengThu, May 14, 2026, 10:59 AM PDT
score 9.2
Single token enables both visual reasoning modes without extra computation
Original: ATLAS: Agentic or Latent Visual Reasoning? One Word is Enough for Both
Source: arxiv.org ↗
Writing ELI5 summary…