Progress-Bench - Human Activities NSE: leaderboard
The in-the-wild set of Progress-Bench, from PROGRESSLM (Zhang et al., 2026): 587 top-down recordings of people doing manipulation tasks such as pushing, pick-and-place, stacking and peg moves. Normalized score error of the model's progress estimate (0-100%) on answerable samples, as the mean of the key-frame and step-text demonstration results; percent, lower is better; direct predictions.
Metric: Normalized score error (%; pointwise error of the predicted progress score; 587 in-the-wild human-activity samples, mean of vision- and text-demonstration results; direct prediction). Source: arxiv.org. Saturation forecast: Around February 2027. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 2.5 VL 7B | 29.2 | #1796 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=progress-bench-human-activities-nse · How It Works · Data refreshed daily, snapshot 2026-11-12.