WatchAct - Moment: leaderboard
The Moment task of WatchAct (Li et al., 2026): after watching a real video of a person in a tabletop scene, the robot planner localizes a described moment in the video and retrieves the information needed to act. A vision-language model writes a high-level Pick/Place/Open/Close plan that is executed symbolically in an aligned LIBERO scene; plan success rate in percent over 135 instances, higher is better.
Metric: Plan success rate (%; the VLM's high-level Pick/Place/Open/Close plan, predicted from a real human video and an instruction, run through a symbolic executor and checked against the task's success predicate; 135 instances). Source: arxiv.org. Saturation forecast: Around September 2027. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Flash | 23.1 | #159 |
| 2 | Gemini 3.1 Pro (Preview) | 22.7 | #91 |
| 3 | Qwen 3 VL 235B A22B (Thinking) | 12.1 | #451 |
| 4 | GPT-5.1 | 7.4 | #225 |
| 5 | Qwen 3 VL 235B A22B Instruct | 5.9 | #521 |
| 6 | Gemini 3.1 Flash Lite | 4.6 | #363 |
| 7 | GPT-5.4 | 4.5 | #121 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=watchact-moment · How It Works · Data refreshed daily, snapshot 2026-11-12.