HELM Robo-Reward-Bench - Roboturk: leaderboard
The "Roboturk" 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: Around September 2027. 25 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 Mini (2025-08-07) | 0.51 | #175 |
| 2 | GPT-5.2 | 0.56 | #114 |
| 3 | GPT-5.5 (2026-04-23) | 0.56 | #23 |
| 4 | GPT-5.5 (xHigh) | 0.59 | #30 (GPT-5.5) |
| 5 | GPT-5.1 (2025-11-13) | 0.62 | #106 |
| 6 | GPT-5 | 0.63 | #101 |
| 7 | GPT-5 Nano | 0.81 | #430 |
| 8 | Gemini 2.5 Flash | 0.98 | #247 |
| 9 | Gemini 3 Pro (Preview) | 0.99 | #67 |
| 10 | Llama 4 Maverick Instruct | 0.99 | #460 |
| 11 | Gemini 2.5 Flash Lite | 1.02 | #426 |
| 12 | Gemini 3 Flash (Preview) | 1.13 | #86 |
| 13 | Gemini 2.5 Pro | 1.31 | #155 |
| 14 | Llama 4 Scout Instruct | 1.35 | #541 |
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-roboturk · How It Works · Data refreshed daily, snapshot 2026-10-24.