AHA-Memes - Fine-Grained Hate Types: leaderboard
Fine-grained classification on AHA-Memes (2026): zero-shot macro-F1 in percent over ten multi-label hate and non-hate type labels for the Arabic memes of the 5K gold test split. Higher is better.
Metric: Fine-grained macro-F1 (%) over ten hate and non-hate type labels (multi-label), zero-shot on the 5K human-annotated gold test split of Arabic memes (image with its OCR text); higher is better. Source: arxiv.org. Saturation forecast: Around May 2028. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro | 34 | #266 |
| 2 | GPT-5 | 30.1 | #167 |
| 3 | Gemini 3.5 Flash | 27.1 | #101 |
| 4 | Qwen 3 VL 8B (Thinking) | 19.1 | #887 |
| 5 | Qwen 3 VL 8B Instruct | 17.6 | #981 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=aha-memes-fine-grained-hate-types · How It Works · Data refreshed daily, snapshot 2026-11-13.