Video-HOCA - Fine-grained Recognition (Ontological): leaderboard
Metric: Macro-F1 (%) of choosing the correct anomaly description among four options (three human-reviewed GPT-OSS-120B distractors) on Video-HOCA's ontological anomalies (entity-centred: an object or agent is internally inconsistent, unstable or does what it should not); 16 uniformly sampled frames, deterministic Instruct-mode decoding (temperature 0, 1,024 tokens); higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 20 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.5 27B (Non-reasoning) | 90.2 | #183 (Qwen 3.5 27B) |
| 2 | Gemini 3 Flash | 85.6 | #93 |
| 3 | Qwen 3 VL 32B Instruct | 84.1 | #276 |
| 4 | Qwen 3 VL 8B Instruct | 83.7 | #401 |
| 5 | Gemini 2.5 Flash | 82.8 | #237 |
| 6 | Qwen 3 VL 30B A3B Instruct | 81.6 | #365 |
| 7 | Qwen 3.5 9B (Non-reasoning) | 74.9 | #363 (Qwen 3.5 9B) |
| 8 | Qwen 2.5 VL 7B Instruct | 67.9 | #643 |
| 9 | GPT-4o | 59.3 | #333 |
| 10 | Qwen 3.5 35B A3B (Non-reasoning) | 32.7 | #250 (Qwen 3.5 35B A3B) |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=video-hoca-fine-grained-recognition-ontological · How It Works · Data refreshed daily, snapshot 2026-10-11.