mmWave-QA - Action Recognition (Few-shot CoT): leaderboard
Action-recognition accuracy on mmWave-QA (2026) under few-shot chain-of-thought prompting, over the Phoenix, IWR6843 and IWR1443 radars. Percent, higher is better.
Metric: Accuracy (%; multiple-choice recognition of the performed action from mmWave radar point clouds serialized as text, 860 questions over the Phoenix, IWR6843 and IWR1443 radars and four body-region action categories; few-shot cot prompting, with in-context examples and a rationale before the answer). Source: arxiv.org. Saturation forecast: Estimated already saturated. 4 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 38.37 | #163 |
| 2 | GPT-4o | 36.98 | #741 |
| 3 | Gemini 2.5 Pro | 36.51 | #262 |
| 4 | Gemini 2.5 Flash | 36.05 | #452 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=mmwave-qa-action-recognition-few-shot-cot · How It Works · Data refreshed daily, snapshot 2026-11-12.