PushupBench: leaderboard

Repetition counting on PushupBench (2026), 446 long-form exercise videos of 22 to 117 seconds, sampled at 5 frames per second: the model reports how many repetitions were performed. Share of clips with the exact count, in percent; higher is better.

Metric: Exact match (%): share of clips whose predicted repetition count equals the ground truth, on PushupBench's 446 long-form exercise clips (22-117 s, 11 creators in 10 countries; 27 clips accept several valid counts) sampled at 5 fps up to 112 frames (Claude models capped at 100 frames by the API), 360p frames, greedy decoding, 10 prompt templates; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 20 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Gemini 3 Flash42.1#175
2Gemini 3 Pro39.8#145
3Gemini 2.5 Pro29.9#276
4Qwen 3 VL 32B Instruct11.8#606
5GPT-510.9#178
6Gemini 2.5 Flash10.9#463
7Qwen 3 VL 32B (Thinking)9.9#573
8Qwen 3 VL 30B A3B (Thinking)9.6#719
9Claude Sonnet 4.59.5#261
10Qwen 2.5 VL 7B9.2#1862
11Gemini 2.0 Flash8.9#723
12Qwen 3 VL 4B Instruct8.9#1332
13Qwen 3 VL 235B A22B Instruct8.6#532
14Qwen 3 VL 4B (Thinking)8.2#1334
15Qwen 3 VL 30B A3B Instruct7.6#895

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=pushupbench · How It Works · Data refreshed daily, snapshot 2026-11-21.