MedAraBench: leaderboard
MedAraBench is an Arabic medical question-answering benchmark for text input, built from multiple-choice questions across 19 specialties and five difficulty levels. The pinned metric, Overall Accuracy (%), measures how often a model selects the correct answer in the reported results table. Source: BenchmarkList’s attributed result table.
Metric: Overall Accuracy (%). Source: benchmarklist.com. Saturation forecast: Estimated already saturated. 16 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | O3 | 76.5 | #132 |
| 2 | GPT-5 | 76.4 | #105 |
| 3 | Claude Sonnet 4 (20250514) | 69.4 | #223 |
| 4 | GPT-4.1 | 67.3 | #258 |
| 5 | Gemini 2.0 Flash | 65.4 | #351 |
| 6 | DeepSeek V3 (0324) | 62 | #349 |
| 7 | Qwen Plus | 61.8 | #393 |
| 8 | Llama 3.3 70B Instruct | 54.7 | #541 |
| 9 | MedGemma-4B-IT | 39 | #811 |
| 10 | c4ai-command-r7B-arabic-02-2025 | 38.1 | #727 |
| 11 | Llama 3.1 8B Instruct | 17 | #1017 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=medarabench · How It Works · Data refreshed daily, snapshot 2026-11-05.