FitAQA - Judgement: leaderboard
FitAQA (Tsinghua University, 2026) evaluates multimodal LLMs on fitness action-quality assessment over 30 bodyweight exercises and a 38-error taxonomy. Judgement F1 scores the error-positive class when the model judges whether an exercise aspect is executed correctly, in percent, higher is better.
Metric: Form-error F1 (%; error-positive class, judging whether an exercise aspect is executed correctly; over FitAQA's fitness action-quality videos (30 bodyweight exercises, 38-error taxonomy); judgement task). Source: arxiv.org. Saturation forecast: Around February 2027. 15 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 72.1 | #91 |
| 2 | GPT-5.5 | 57.8 | #48 |
| 3 | GPT-5.4 | 50.5 | #121 |
| 4 | Qwen 3.5 27B (Non-reasoning) | 47.1 | #358 (Qwen 3.5 27B) |
| 5 | Qwen 3.5 9B (Non-reasoning) | 45.7 | #803 (Qwen 3.5 9B) |
| 6 | Qwen 3.5 27B (Thinking) | 42.6 | #358 (Qwen 3.5 27B) |
| 7 | Qwen 3.5 9B (Thinking) | 39.9 | #803 (Qwen 3.5 9B) |
| 8 | Qwen 3.5 35B A3B (Thinking) | 39 | #480 (Qwen 3.5 35B A3B) |
| 9 | Qwen 3.5 35B A3B (Non-reasoning) | 33.4 | #480 (Qwen 3.5 35B A3B) |
| 10 | Gemma 4 31B | 27.8 | #393 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=fitaqa-judgement · How It Works · Data refreshed daily, snapshot 2026-11-12.