WatchAct - Task Continuation: leaderboard
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; 170 instances). Source: arxiv.org. Saturation forecast: Around March 2028. 9 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 52.4 |
| 2 | Qwen 3 VL 235B A22B (Thinking) | 44.1 |
| 3 | GPT-5.4 | 43.5 |
| 4 | Gemini 3 Flash | 42.9 |
| 5 | Gemini 3.1 Flash Lite | 42.9 |
| 6 | GPT-5.1 | 37.1 |
| 7 | Qwen 3 VL 235B A22B Instruct | 27.6 |
Interactive version: theaggregate.ai/benchmark?slug=watchact-task-continuation · How It Works · Data refreshed daily, snapshot 2026-09-26.