ResearchBenchmarksStructured video reasoning on Maze3D HardFollow95.3Exact MatchTBS-2072.10878.12984.1590.171Jun 11, 2026Evaluation ResultsMethodMethodLinksExact MatchTBS-20Budget=20Budget=202026.0695.3TBS-10Budget=10Budget=102026.0694.3BoN-20Budget=20Budget=202026.0693.7BoN-10Budget=10Budget=102026.0692.7BoN-5Budget=5Budget=52026.0691BoN-1Budget=1Budget=12026.0673