WatchAct - Reversal: 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; 225 instances). Source: arxiv.org. Saturation forecast: Around July 2027. 9 models tracked.

Top models

#ModelScore
1Gemini 3.1 Pro (Preview)30.2
2GPT-5.427.1
3GPT-5.122.2
4Gemini 3 Flash21.3
5Qwen 3 VL 235B A22B Instruct8.4
6Qwen 3 VL 235B A22B (Thinking)6.7
7Gemini 3.1 Flash Lite4.9

Interactive version: theaggregate.ai/benchmark?slug=watchact-reversal · How It Works · Data refreshed daily, snapshot 2026-09-26.