HELM Robo-Reward-Bench - Tokyo U Lsmo Converted Externally To Rlds: leaderboard
The "Tokyo U Lsmo Converted Externally To Rlds" 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 | Gemini 2.5 Flash | 0.46 | #247 |
| 2 | Gemini 3 Flash (Preview) | 0.55 | #86 |
| 3 | GPT-5 | 0.65 | #101 |
| 4 | GPT-5 Mini (2025-08-07) | 0.68 | #175 |
| 5 | GPT-5.5 (xHigh) | 0.73 | #30 (GPT-5.5) |
| 6 | GPT-5.5 (2026-04-23) | 0.73 | #23 |
| 7 | GPT-5.1 (2025-11-13) | 0.76 | #106 |
| 8 | Gemini 2.5 Pro | 0.92 | #155 |
| 9 | Gemini 3 Pro (Preview) | 0.97 | #67 |
| 10 | GPT-5.2 | 0.99 | #114 |
| 11 | Gemini 2.5 Flash Lite | 1.2 | #426 |
| 12 | GPT-5 Nano | 1.37 | #430 |
| 13 | Llama 4 Maverick Instruct | 1.61 | #460 |
| 14 | Llama 4 Scout Instruct | 1.75 | #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-tokyo-u-lsmo-converted-externally-to-rlds · How It Works · Data refreshed daily, snapshot 2026-10-24.