Progress-Bench: leaderboard
Progress-Bench, from PROGRESSLM (Zhang et al., 2026), asks vision-language models to estimate how far a robot manipulation task has progressed (0-100%) from one observation and a demonstration, over 3,325 observations from 240 RoboMind trajectories. Progress rank correlation is the Spearman correlation (x100) between predicted and true progress along each trajectory, averaged over trajectories and micro-averaged over key-frame and step-text demonstrations; higher is better, from -100 to 100. Scores are direct predictions, made without prompted reasoning.
Metric: Progress rank correlation (Spearman x 100 between predicted and true progress along each trajectory, averaged over trajectories; answerable items with vision- and text-based demonstrations, micro-averaged; direct prediction). Source: arxiv.org. Saturation forecast: Estimated already saturated. 12 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 82.5 | #163 |
| 2 | GPT-5 Mini | 81.1 | #349 |
| 3 | Qwen 2.5 VL 7B | 31 | #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 · How It Works · Data refreshed daily, snapshot 2026-11-12.