HELM Robo-Reward-Bench - Robo Reward Bench Preference Ranking: leaderboard
The "Robo Reward Bench Preference Ranking" 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: EM. Source: crfm.stanford.edu. Saturation forecast: Estimated already saturated. 18 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4.1 (2025-04-14) | 67.5 | #250 |
| 2 | GPT-4.5 | 65.5 | #276 |
| 3 | Gemini 2.5 Flash (Preview 04-17) | 62.5 | #235 |
| 4 | Gemini 2.5 Flash (Preview 05-20) | 59 | #214 |
| 5 | O1 (2024-12-17) | 58.5 | #152 |
| 6 | GPT-4o (2024-11-20) | 55 | #378 |
| 7 | Gemini 2.0 Flash Lite | 53.5 | #453 |
| 8 | Gemini 2.5 Pro (Preview 05-06) | 52 | #148 |
| 9 | GPT-4.1 Mini | 49 | #351 |
| 10 | Gemini 2.0 Flash | 49 | #344 |
| 11 | Qwen2.5-Omni-7B | 38.5 | #632 |
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-preference-ranking · How It Works · Data refreshed daily, snapshot 2026-10-23.