BloomBench - Arabic - Evaluate: leaderboard
BloomBench (Almieyar-Oryx series, QCRI, 2026) is a bilingual English and Arabic multiple-choice benchmark for vision-language models organized by the six levels of the revised Bloom taxonomy, scored as zero-shot accuracy in percent from the answer letter parsed from generated text (chance 25). This board is accuracy on the 592 Modern Standard Arabic items of the Evaluate level (judging logical coherence, harm and safety, and image quality); higher is better.
Metric: Accuracy (%) on the 592 Arabic BloomBench items at the Bloom level Evaluate (judging coherence, harm and safety, and image quality), four-option image-question multiple choice, zero-shot, temperature 0, the chosen letter parsed from the generated answer (Regex-based Answer Extraction; unparseable answers count as wrong); chance 25; higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemma 4 31B | 93 | #394 |
| 2 | Gemma 3 27B | 89 | #1631 |
| 3 | Gemma 4 26B A4B | 89 | #615 |
| 4 | Gemma 3 12B | 88 | #2000 |
| 5 | GPT-4o Mini | 83 | #1623 |
| 6 | Qwen 2.5 VL 7B | 82 | #1787 |
| 7 | Qwen 2 VL 7B | 79 | #2261 |
| 8 | Gemma 3 4B | 73 | #5064 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=bloombench-arabic-evaluate · How It Works · Data refreshed daily, snapshot 2026-11-13.