D&A-Bench - D&A protocol: leaderboard
D&A TableII: D&A-Bench human-ranked robot trajectories; Kendall rank agreement with human ordering, theoretical-1 to 1. Exact named D&A prompting configurations using GPT-6 Astra and Gemini 3Flash, same native benchmark, retained separately from naive-prompt/othertrained reward-model baselines. Reported Gemini 3Flash 0.672 vsGPT 6Astra 0.790. No effort setting is disclosed. RoboReward comparisons explicitly cite external baselines and are not pooled; figure-only ablations retain separate review action. Metric: Kendall rank correlation.
Metric: Kendall rank correlation. Source: arxiv.org. Saturation forecast: Around February 2027. 2 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-6 Astra | 0.79 | #6 |
No result here: #3 Claude Opus 5.5, #7 GPT-6.1 Sol, #9 Gemini 4 Argon.
Interactive version: theaggregate.ai/benchmark?slug=d-a-bench-d-a-protocol · How It Works · Data refreshed daily, snapshot 2026-11-05.