Progress-Bench - NSE: leaderboard
Normalized score error of Progress-Bench, from PROGRESSLM (Zhang et al., 2026): the pointwise error of a model's estimate of how far a robot manipulation task has progressed (0-100%), over answerable observations with key-frame or step-text demonstrations, micro-averaged. Percent, lower is better; direct predictions without prompted reasoning.
Metric: Normalized score error (%; pointwise error of the predicted progress score; answerable items with vision- and text-based demonstrations, micro-averaged; direct prediction). Source: arxiv.org. Saturation forecast: Estimated already saturated. 13 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 19.9 | #163 |
| 2 | GPT-5 Mini | 20.8 | #349 |
| 3 | Qwen 2.5 VL 7B | 35 | #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-nse · How It Works · Data refreshed daily, snapshot 2026-11-12.