MORPHOGEN - Arabic Corpus Gender Accuracy: leaderboard
Metric: Corpus-level gender accuracy (%): correctly generated gendered terms over all reference gendered terms of the test set, in the GENFORM task of MORPHOGEN: rewrite a first-person Arabic sentence for a speaker of the opposite grammatical gender while keeping its meaning (both directions, human-corrected references), zero-shot with the model chat template, temperature 0.1; higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 15 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-4o Mini | 80.27 |
| 2 | Gemma 3 27B | 74.74 |
| 3 | Phi-4 | 66.15 |
| 4 | Gemma 3 12B | 65.52 |
| 5 | Llama 3.3 70B | 64.37 |
| 6 | Qwen 3 14B | 56.08 |
| 7 | Qwen 3 32B | 53 |
| 8 | Qwen 3 8B | 51.01 |
| 9 | Gemma 3 4B | 48.93 |
| 10 | Llama 3.1 8B | 45.51 |
| 11 | Gemma 2 9B | 45.26 |
| 12 | Qwen 3 4B | 35.97 |
| 13 | Llama 3.2 3B | 17.75 |
| 14 | Gemma 2 2B | 14.1 |
| 15 | Qwen 2.5 0.5B | 4.59 |
Interactive version: theaggregate.ai/benchmark?slug=morphogen-arabic-corpus-gender-accuracy · How It Works · Data refreshed daily, snapshot 2026-10-07.