Pith. sign in

← back to paper

Review history

arxiv: 2608.01207 · 2 revisions

It's the Decoding Format, Not the Perturbation: Auditing Consistency-Based Selection for Vision-Language Test-Time Scaling

  1. 2026-08-06 CONDITIONAL HIGH v1.4.0-alltime-deepseek-medium novelty 7.0
    131828 ms 17407 in 12648 out 2026-08-06T04:12:04.420915+00:00
  2. 2026-08-06 CONDITIONAL MODERATE v1.4.0-alltime-deepseek-medium novelty 6.0
    164716 ms 13400 in 17103 out 2026-08-06T00:26:45.524811+00:00