HELM Robo-Reward-Bench - Robo Reward Bench Progress Prediction: leaderboard
The "Robo Reward Bench Progress Prediction" subtask of HELM Robo-Reward-Bench. Stanford HELM's Robo-Reward-Bench: vision-language models act as reward models over robot manipulation episodes, predicting task progress or success from video.
Metric: Absolute error. Source: crfm.stanford.edu. Saturation forecast: Estimated already saturated. 18 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro (Preview 05-06) | 18.15 | #149 |
| 2 | GPT-4.1 (2025-04-14) | 20.46 | #252 |
| 3 | O1 (2024-12-17) | 22.2 | #153 |
| 4 | GPT-4o (2024-11-20) | 22.89 | #379 |
| 5 | GPT-4.5 | 24.67 | #276 |
| 6 | Gemini 2.5 Flash (Preview 04-17) | 25.1 | #236 |
| 7 | Gemini 2.5 Flash (Preview 05-20) | 26.56 | #217 |
| 8 | Gemini 2.0 Flash Lite | 27.54 | #455 |
| 9 | Gemini 2.0 Flash | 27.63 | #345 |
| 10 | GPT-4.1 Mini | 36.17 | #358 |
| 11 | Qwen2.5-Omni-7B | 40.45 | #634 |
No result here: #4 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=helm-robo-reward-bench-robo-reward-bench-progress-prediction · How It Works · Data refreshed daily, snapshot 2026-10-24.